2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 static struct kmem_cache *btrfs_delalloc_work_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_transaction_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
91 static int btrfs_setsize(struct inode *inode, loff_t newsize);
92 static int btrfs_truncate(struct inode *inode);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, int *page_started,
97 unsigned long *nr_written, int unlock);
98 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
99 u64 len, u64 orig_start,
100 u64 block_start, u64 block_len,
101 u64 orig_block_len, int type);
103 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
104 struct inode *inode, struct inode *dir,
105 const struct qstr *qstr)
109 err = btrfs_init_acl(trans, inode, dir);
111 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
116 * this does all the hard work for inserting an inline extent into
117 * the btree. The caller should have done a btrfs_drop_extents so that
118 * no overlapping inline items exist in the btree
120 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
121 struct btrfs_root *root, struct inode *inode,
122 u64 start, size_t size, size_t compressed_size,
124 struct page **compressed_pages)
126 struct btrfs_key key;
127 struct btrfs_path *path;
128 struct extent_buffer *leaf;
129 struct page *page = NULL;
132 struct btrfs_file_extent_item *ei;
135 size_t cur_size = size;
137 unsigned long offset;
139 if (compressed_size && compressed_pages)
140 cur_size = compressed_size;
142 path = btrfs_alloc_path();
146 path->leave_spinning = 1;
148 key.objectid = btrfs_ino(inode);
150 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
151 datasize = btrfs_file_extent_calc_inline_size(cur_size);
153 inode_add_bytes(inode, size);
154 ret = btrfs_insert_empty_item(trans, root, path, &key,
160 leaf = path->nodes[0];
161 ei = btrfs_item_ptr(leaf, path->slots[0],
162 struct btrfs_file_extent_item);
163 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
164 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
165 btrfs_set_file_extent_encryption(leaf, ei, 0);
166 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
167 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
168 ptr = btrfs_file_extent_inline_start(ei);
170 if (compress_type != BTRFS_COMPRESS_NONE) {
173 while (compressed_size > 0) {
174 cpage = compressed_pages[i];
175 cur_size = min_t(unsigned long, compressed_size,
178 kaddr = kmap_atomic(cpage);
179 write_extent_buffer(leaf, kaddr, ptr, cur_size);
180 kunmap_atomic(kaddr);
184 compressed_size -= cur_size;
186 btrfs_set_file_extent_compression(leaf, ei,
189 page = find_get_page(inode->i_mapping,
190 start >> PAGE_CACHE_SHIFT);
191 btrfs_set_file_extent_compression(leaf, ei, 0);
192 kaddr = kmap_atomic(page);
193 offset = start & (PAGE_CACHE_SIZE - 1);
194 write_extent_buffer(leaf, kaddr + offset, ptr, size);
195 kunmap_atomic(kaddr);
196 page_cache_release(page);
198 btrfs_mark_buffer_dirty(leaf);
199 btrfs_free_path(path);
202 * we're an inline extent, so nobody can
203 * extend the file past i_size without locking
204 * a page we already have locked.
206 * We must do any isize and inode updates
207 * before we unlock the pages. Otherwise we
208 * could end up racing with unlink.
210 BTRFS_I(inode)->disk_i_size = inode->i_size;
211 ret = btrfs_update_inode(trans, root, inode);
215 btrfs_free_path(path);
221 * conditionally insert an inline extent into the file. This
222 * does the checks required to make sure the data is small enough
223 * to fit as an inline extent.
225 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
226 struct btrfs_root *root,
227 struct inode *inode, u64 start, u64 end,
228 size_t compressed_size, int compress_type,
229 struct page **compressed_pages)
231 u64 isize = i_size_read(inode);
232 u64 actual_end = min(end + 1, isize);
233 u64 inline_len = actual_end - start;
234 u64 aligned_end = (end + root->sectorsize - 1) &
235 ~((u64)root->sectorsize - 1);
236 u64 data_len = inline_len;
240 data_len = compressed_size;
243 actual_end >= PAGE_CACHE_SIZE ||
244 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
246 (actual_end & (root->sectorsize - 1)) == 0) ||
248 data_len > root->fs_info->max_inline) {
252 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
256 if (isize > actual_end)
257 inline_len = min_t(u64, isize, actual_end);
258 ret = insert_inline_extent(trans, root, inode, start,
259 inline_len, compressed_size,
260 compress_type, compressed_pages);
261 if (ret && ret != -ENOSPC) {
262 btrfs_abort_transaction(trans, root, ret);
264 } else if (ret == -ENOSPC) {
268 btrfs_delalloc_release_metadata(inode, end + 1 - start);
269 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
273 struct async_extent {
278 unsigned long nr_pages;
280 struct list_head list;
285 struct btrfs_root *root;
286 struct page *locked_page;
289 struct list_head extents;
290 struct btrfs_work work;
293 static noinline int add_async_extent(struct async_cow *cow,
294 u64 start, u64 ram_size,
297 unsigned long nr_pages,
300 struct async_extent *async_extent;
302 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
303 BUG_ON(!async_extent); /* -ENOMEM */
304 async_extent->start = start;
305 async_extent->ram_size = ram_size;
306 async_extent->compressed_size = compressed_size;
307 async_extent->pages = pages;
308 async_extent->nr_pages = nr_pages;
309 async_extent->compress_type = compress_type;
310 list_add_tail(&async_extent->list, &cow->extents);
315 * we create compressed extents in two phases. The first
316 * phase compresses a range of pages that have already been
317 * locked (both pages and state bits are locked).
319 * This is done inside an ordered work queue, and the compression
320 * is spread across many cpus. The actual IO submission is step
321 * two, and the ordered work queue takes care of making sure that
322 * happens in the same order things were put onto the queue by
323 * writepages and friends.
325 * If this code finds it can't get good compression, it puts an
326 * entry onto the work queue to write the uncompressed bytes. This
327 * makes sure that both compressed inodes and uncompressed inodes
328 * are written in the same order that the flusher thread sent them
331 static noinline int compress_file_range(struct inode *inode,
332 struct page *locked_page,
334 struct async_cow *async_cow,
337 struct btrfs_root *root = BTRFS_I(inode)->root;
338 struct btrfs_trans_handle *trans;
340 u64 blocksize = root->sectorsize;
342 u64 isize = i_size_read(inode);
344 struct page **pages = NULL;
345 unsigned long nr_pages;
346 unsigned long nr_pages_ret = 0;
347 unsigned long total_compressed = 0;
348 unsigned long total_in = 0;
349 unsigned long max_compressed = 128 * 1024;
350 unsigned long max_uncompressed = 128 * 1024;
353 int compress_type = root->fs_info->compress_type;
355 /* if this is a small write inside eof, kick off a defrag */
356 if ((end - start + 1) < 16 * 1024 &&
357 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
358 btrfs_add_inode_defrag(NULL, inode);
360 actual_end = min_t(u64, isize, end + 1);
363 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
364 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
367 * we don't want to send crud past the end of i_size through
368 * compression, that's just a waste of CPU time. So, if the
369 * end of the file is before the start of our current
370 * requested range of bytes, we bail out to the uncompressed
371 * cleanup code that can deal with all of this.
373 * It isn't really the fastest way to fix things, but this is a
374 * very uncommon corner.
376 if (actual_end <= start)
377 goto cleanup_and_bail_uncompressed;
379 total_compressed = actual_end - start;
381 /* we want to make sure that amount of ram required to uncompress
382 * an extent is reasonable, so we limit the total size in ram
383 * of a compressed extent to 128k. This is a crucial number
384 * because it also controls how easily we can spread reads across
385 * cpus for decompression.
387 * We also want to make sure the amount of IO required to do
388 * a random read is reasonably small, so we limit the size of
389 * a compressed extent to 128k.
391 total_compressed = min(total_compressed, max_uncompressed);
392 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
393 num_bytes = max(blocksize, num_bytes);
398 * we do compression for mount -o compress and when the
399 * inode has not been flagged as nocompress. This flag can
400 * change at any time if we discover bad compression ratios.
402 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
403 (btrfs_test_opt(root, COMPRESS) ||
404 (BTRFS_I(inode)->force_compress) ||
405 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
407 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
409 /* just bail out to the uncompressed code */
413 if (BTRFS_I(inode)->force_compress)
414 compress_type = BTRFS_I(inode)->force_compress;
416 ret = btrfs_compress_pages(compress_type,
417 inode->i_mapping, start,
418 total_compressed, pages,
419 nr_pages, &nr_pages_ret,
425 unsigned long offset = total_compressed &
426 (PAGE_CACHE_SIZE - 1);
427 struct page *page = pages[nr_pages_ret - 1];
430 /* zero the tail end of the last page, we might be
431 * sending it down to disk
434 kaddr = kmap_atomic(page);
435 memset(kaddr + offset, 0,
436 PAGE_CACHE_SIZE - offset);
437 kunmap_atomic(kaddr);
444 trans = btrfs_join_transaction(root);
446 ret = PTR_ERR(trans);
448 goto cleanup_and_out;
450 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
452 /* lets try to make an inline extent */
453 if (ret || total_in < (actual_end - start)) {
454 /* we didn't compress the entire range, try
455 * to make an uncompressed inline extent.
457 ret = cow_file_range_inline(trans, root, inode,
458 start, end, 0, 0, NULL);
460 /* try making a compressed inline extent */
461 ret = cow_file_range_inline(trans, root, inode,
464 compress_type, pages);
468 * inline extent creation worked or returned error,
469 * we don't need to create any more async work items.
470 * Unlock and free up our temp pages.
472 extent_clear_unlock_delalloc(inode,
473 &BTRFS_I(inode)->io_tree,
475 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
476 EXTENT_CLEAR_DELALLOC |
477 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
479 btrfs_end_transaction(trans, root);
482 btrfs_end_transaction(trans, root);
487 * we aren't doing an inline extent round the compressed size
488 * up to a block size boundary so the allocator does sane
491 total_compressed = (total_compressed + blocksize - 1) &
495 * one last check to make sure the compression is really a
496 * win, compare the page count read with the blocks on disk
498 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
499 ~(PAGE_CACHE_SIZE - 1);
500 if (total_compressed >= total_in) {
503 num_bytes = total_in;
506 if (!will_compress && pages) {
508 * the compression code ran but failed to make things smaller,
509 * free any pages it allocated and our page pointer array
511 for (i = 0; i < nr_pages_ret; i++) {
512 WARN_ON(pages[i]->mapping);
513 page_cache_release(pages[i]);
517 total_compressed = 0;
520 /* flag the file so we don't compress in the future */
521 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
522 !(BTRFS_I(inode)->force_compress)) {
523 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
529 /* the async work queues will take care of doing actual
530 * allocation on disk for these compressed pages,
531 * and will submit them to the elevator.
533 add_async_extent(async_cow, start, num_bytes,
534 total_compressed, pages, nr_pages_ret,
537 if (start + num_bytes < end) {
544 cleanup_and_bail_uncompressed:
546 * No compression, but we still need to write the pages in
547 * the file we've been given so far. redirty the locked
548 * page if it corresponds to our extent and set things up
549 * for the async work queue to run cow_file_range to do
550 * the normal delalloc dance
552 if (page_offset(locked_page) >= start &&
553 page_offset(locked_page) <= end) {
554 __set_page_dirty_nobuffers(locked_page);
555 /* unlocked later on in the async handlers */
557 add_async_extent(async_cow, start, end - start + 1,
558 0, NULL, 0, BTRFS_COMPRESS_NONE);
566 for (i = 0; i < nr_pages_ret; i++) {
567 WARN_ON(pages[i]->mapping);
568 page_cache_release(pages[i]);
575 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
577 EXTENT_CLEAR_UNLOCK_PAGE |
579 EXTENT_CLEAR_DELALLOC |
580 EXTENT_SET_WRITEBACK |
581 EXTENT_END_WRITEBACK);
582 if (!trans || IS_ERR(trans))
583 btrfs_error(root->fs_info, ret, "Failed to join transaction");
585 btrfs_abort_transaction(trans, root, ret);
590 * phase two of compressed writeback. This is the ordered portion
591 * of the code, which only gets called in the order the work was
592 * queued. We walk all the async extents created by compress_file_range
593 * and send them down to the disk.
595 static noinline int submit_compressed_extents(struct inode *inode,
596 struct async_cow *async_cow)
598 struct async_extent *async_extent;
600 struct btrfs_trans_handle *trans;
601 struct btrfs_key ins;
602 struct extent_map *em;
603 struct btrfs_root *root = BTRFS_I(inode)->root;
604 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
605 struct extent_io_tree *io_tree;
608 if (list_empty(&async_cow->extents))
612 while (!list_empty(&async_cow->extents)) {
613 async_extent = list_entry(async_cow->extents.next,
614 struct async_extent, list);
615 list_del(&async_extent->list);
617 io_tree = &BTRFS_I(inode)->io_tree;
620 /* did the compression code fall back to uncompressed IO? */
621 if (!async_extent->pages) {
622 int page_started = 0;
623 unsigned long nr_written = 0;
625 lock_extent(io_tree, async_extent->start,
626 async_extent->start +
627 async_extent->ram_size - 1);
629 /* allocate blocks */
630 ret = cow_file_range(inode, async_cow->locked_page,
632 async_extent->start +
633 async_extent->ram_size - 1,
634 &page_started, &nr_written, 0);
639 * if page_started, cow_file_range inserted an
640 * inline extent and took care of all the unlocking
641 * and IO for us. Otherwise, we need to submit
642 * all those pages down to the drive.
644 if (!page_started && !ret)
645 extent_write_locked_range(io_tree,
646 inode, async_extent->start,
647 async_extent->start +
648 async_extent->ram_size - 1,
656 lock_extent(io_tree, async_extent->start,
657 async_extent->start + async_extent->ram_size - 1);
659 trans = btrfs_join_transaction(root);
661 ret = PTR_ERR(trans);
663 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
664 ret = btrfs_reserve_extent(trans, root,
665 async_extent->compressed_size,
666 async_extent->compressed_size,
667 0, alloc_hint, &ins, 1);
668 if (ret && ret != -ENOSPC)
669 btrfs_abort_transaction(trans, root, ret);
670 btrfs_end_transaction(trans, root);
675 for (i = 0; i < async_extent->nr_pages; i++) {
676 WARN_ON(async_extent->pages[i]->mapping);
677 page_cache_release(async_extent->pages[i]);
679 kfree(async_extent->pages);
680 async_extent->nr_pages = 0;
681 async_extent->pages = NULL;
682 unlock_extent(io_tree, async_extent->start,
683 async_extent->start +
684 async_extent->ram_size - 1);
687 goto out_free; /* JDM: Requeue? */
691 * here we're doing allocation and writeback of the
694 btrfs_drop_extent_cache(inode, async_extent->start,
695 async_extent->start +
696 async_extent->ram_size - 1, 0);
698 em = alloc_extent_map();
699 BUG_ON(!em); /* -ENOMEM */
700 em->start = async_extent->start;
701 em->len = async_extent->ram_size;
702 em->orig_start = em->start;
704 em->block_start = ins.objectid;
705 em->block_len = ins.offset;
706 em->orig_block_len = ins.offset;
707 em->bdev = root->fs_info->fs_devices->latest_bdev;
708 em->compress_type = async_extent->compress_type;
709 set_bit(EXTENT_FLAG_PINNED, &em->flags);
710 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
714 write_lock(&em_tree->lock);
715 ret = add_extent_mapping(em_tree, em);
718 &em_tree->modified_extents);
719 write_unlock(&em_tree->lock);
720 if (ret != -EEXIST) {
724 btrfs_drop_extent_cache(inode, async_extent->start,
725 async_extent->start +
726 async_extent->ram_size - 1, 0);
729 ret = btrfs_add_ordered_extent_compress(inode,
732 async_extent->ram_size,
734 BTRFS_ORDERED_COMPRESSED,
735 async_extent->compress_type);
736 BUG_ON(ret); /* -ENOMEM */
739 * clear dirty, set writeback and unlock the pages.
741 extent_clear_unlock_delalloc(inode,
742 &BTRFS_I(inode)->io_tree,
744 async_extent->start +
745 async_extent->ram_size - 1,
746 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
747 EXTENT_CLEAR_UNLOCK |
748 EXTENT_CLEAR_DELALLOC |
749 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
751 ret = btrfs_submit_compressed_write(inode,
753 async_extent->ram_size,
755 ins.offset, async_extent->pages,
756 async_extent->nr_pages);
758 BUG_ON(ret); /* -ENOMEM */
759 alloc_hint = ins.objectid + ins.offset;
771 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
774 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
775 struct extent_map *em;
778 read_lock(&em_tree->lock);
779 em = search_extent_mapping(em_tree, start, num_bytes);
782 * if block start isn't an actual block number then find the
783 * first block in this inode and use that as a hint. If that
784 * block is also bogus then just don't worry about it.
786 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
788 em = search_extent_mapping(em_tree, 0, 0);
789 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
790 alloc_hint = em->block_start;
794 alloc_hint = em->block_start;
798 read_unlock(&em_tree->lock);
804 * when extent_io.c finds a delayed allocation range in the file,
805 * the call backs end up in this code. The basic idea is to
806 * allocate extents on disk for the range, and create ordered data structs
807 * in ram to track those extents.
809 * locked_page is the page that writepage had locked already. We use
810 * it to make sure we don't do extra locks or unlocks.
812 * *page_started is set to one if we unlock locked_page and do everything
813 * required to start IO on it. It may be clean and already done with
816 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
818 struct btrfs_root *root,
819 struct page *locked_page,
820 u64 start, u64 end, int *page_started,
821 unsigned long *nr_written,
826 unsigned long ram_size;
829 u64 blocksize = root->sectorsize;
830 struct btrfs_key ins;
831 struct extent_map *em;
832 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
835 BUG_ON(btrfs_is_free_space_inode(inode));
837 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
838 num_bytes = max(blocksize, num_bytes);
839 disk_num_bytes = num_bytes;
841 /* if this is a small write inside eof, kick off defrag */
842 if (num_bytes < 64 * 1024 &&
843 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
844 btrfs_add_inode_defrag(trans, inode);
847 /* lets try to make an inline extent */
848 ret = cow_file_range_inline(trans, root, inode,
849 start, end, 0, 0, NULL);
851 extent_clear_unlock_delalloc(inode,
852 &BTRFS_I(inode)->io_tree,
854 EXTENT_CLEAR_UNLOCK_PAGE |
855 EXTENT_CLEAR_UNLOCK |
856 EXTENT_CLEAR_DELALLOC |
858 EXTENT_SET_WRITEBACK |
859 EXTENT_END_WRITEBACK);
861 *nr_written = *nr_written +
862 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
865 } else if (ret < 0) {
866 btrfs_abort_transaction(trans, root, ret);
871 BUG_ON(disk_num_bytes >
872 btrfs_super_total_bytes(root->fs_info->super_copy));
874 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
875 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
877 while (disk_num_bytes > 0) {
880 cur_alloc_size = disk_num_bytes;
881 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
882 root->sectorsize, 0, alloc_hint,
885 btrfs_abort_transaction(trans, root, ret);
889 em = alloc_extent_map();
890 BUG_ON(!em); /* -ENOMEM */
892 em->orig_start = em->start;
893 ram_size = ins.offset;
894 em->len = ins.offset;
896 em->block_start = ins.objectid;
897 em->block_len = ins.offset;
898 em->orig_block_len = ins.offset;
899 em->bdev = root->fs_info->fs_devices->latest_bdev;
900 set_bit(EXTENT_FLAG_PINNED, &em->flags);
904 write_lock(&em_tree->lock);
905 ret = add_extent_mapping(em_tree, em);
908 &em_tree->modified_extents);
909 write_unlock(&em_tree->lock);
910 if (ret != -EEXIST) {
914 btrfs_drop_extent_cache(inode, start,
915 start + ram_size - 1, 0);
918 cur_alloc_size = ins.offset;
919 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
920 ram_size, cur_alloc_size, 0);
921 BUG_ON(ret); /* -ENOMEM */
923 if (root->root_key.objectid ==
924 BTRFS_DATA_RELOC_TREE_OBJECTID) {
925 ret = btrfs_reloc_clone_csums(inode, start,
928 btrfs_abort_transaction(trans, root, ret);
933 if (disk_num_bytes < cur_alloc_size)
936 /* we're not doing compressed IO, don't unlock the first
937 * page (which the caller expects to stay locked), don't
938 * clear any dirty bits and don't set any writeback bits
940 * Do set the Private2 bit so we know this page was properly
941 * setup for writepage
943 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
944 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
947 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
948 start, start + ram_size - 1,
950 disk_num_bytes -= cur_alloc_size;
951 num_bytes -= cur_alloc_size;
952 alloc_hint = ins.objectid + ins.offset;
953 start += cur_alloc_size;
959 extent_clear_unlock_delalloc(inode,
960 &BTRFS_I(inode)->io_tree,
961 start, end, locked_page,
962 EXTENT_CLEAR_UNLOCK_PAGE |
963 EXTENT_CLEAR_UNLOCK |
964 EXTENT_CLEAR_DELALLOC |
966 EXTENT_SET_WRITEBACK |
967 EXTENT_END_WRITEBACK);
972 static noinline int cow_file_range(struct inode *inode,
973 struct page *locked_page,
974 u64 start, u64 end, int *page_started,
975 unsigned long *nr_written,
978 struct btrfs_trans_handle *trans;
979 struct btrfs_root *root = BTRFS_I(inode)->root;
982 trans = btrfs_join_transaction(root);
984 extent_clear_unlock_delalloc(inode,
985 &BTRFS_I(inode)->io_tree,
986 start, end, locked_page,
987 EXTENT_CLEAR_UNLOCK_PAGE |
988 EXTENT_CLEAR_UNLOCK |
989 EXTENT_CLEAR_DELALLOC |
991 EXTENT_SET_WRITEBACK |
992 EXTENT_END_WRITEBACK);
993 return PTR_ERR(trans);
995 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
997 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
998 page_started, nr_written, unlock);
1000 btrfs_end_transaction(trans, root);
1006 * work queue call back to started compression on a file and pages
1008 static noinline void async_cow_start(struct btrfs_work *work)
1010 struct async_cow *async_cow;
1012 async_cow = container_of(work, struct async_cow, work);
1014 compress_file_range(async_cow->inode, async_cow->locked_page,
1015 async_cow->start, async_cow->end, async_cow,
1017 if (num_added == 0) {
1018 btrfs_add_delayed_iput(async_cow->inode);
1019 async_cow->inode = NULL;
1024 * work queue call back to submit previously compressed pages
1026 static noinline void async_cow_submit(struct btrfs_work *work)
1028 struct async_cow *async_cow;
1029 struct btrfs_root *root;
1030 unsigned long nr_pages;
1032 async_cow = container_of(work, struct async_cow, work);
1034 root = async_cow->root;
1035 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1038 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1040 waitqueue_active(&root->fs_info->async_submit_wait))
1041 wake_up(&root->fs_info->async_submit_wait);
1043 if (async_cow->inode)
1044 submit_compressed_extents(async_cow->inode, async_cow);
1047 static noinline void async_cow_free(struct btrfs_work *work)
1049 struct async_cow *async_cow;
1050 async_cow = container_of(work, struct async_cow, work);
1051 if (async_cow->inode)
1052 btrfs_add_delayed_iput(async_cow->inode);
1056 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1057 u64 start, u64 end, int *page_started,
1058 unsigned long *nr_written)
1060 struct async_cow *async_cow;
1061 struct btrfs_root *root = BTRFS_I(inode)->root;
1062 unsigned long nr_pages;
1064 int limit = 10 * 1024 * 1024;
1066 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1067 1, 0, NULL, GFP_NOFS);
1068 while (start < end) {
1069 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1070 BUG_ON(!async_cow); /* -ENOMEM */
1071 async_cow->inode = igrab(inode);
1072 async_cow->root = root;
1073 async_cow->locked_page = locked_page;
1074 async_cow->start = start;
1076 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1079 cur_end = min(end, start + 512 * 1024 - 1);
1081 async_cow->end = cur_end;
1082 INIT_LIST_HEAD(&async_cow->extents);
1084 async_cow->work.func = async_cow_start;
1085 async_cow->work.ordered_func = async_cow_submit;
1086 async_cow->work.ordered_free = async_cow_free;
1087 async_cow->work.flags = 0;
1089 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1091 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1093 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1096 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1097 wait_event(root->fs_info->async_submit_wait,
1098 (atomic_read(&root->fs_info->async_delalloc_pages) <
1102 while (atomic_read(&root->fs_info->async_submit_draining) &&
1103 atomic_read(&root->fs_info->async_delalloc_pages)) {
1104 wait_event(root->fs_info->async_submit_wait,
1105 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1109 *nr_written += nr_pages;
1110 start = cur_end + 1;
1116 static noinline int csum_exist_in_range(struct btrfs_root *root,
1117 u64 bytenr, u64 num_bytes)
1120 struct btrfs_ordered_sum *sums;
1123 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1124 bytenr + num_bytes - 1, &list, 0);
1125 if (ret == 0 && list_empty(&list))
1128 while (!list_empty(&list)) {
1129 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1130 list_del(&sums->list);
1137 * when nowcow writeback call back. This checks for snapshots or COW copies
1138 * of the extents that exist in the file, and COWs the file as required.
1140 * If no cow copies or snapshots exist, we write directly to the existing
1143 static noinline int run_delalloc_nocow(struct inode *inode,
1144 struct page *locked_page,
1145 u64 start, u64 end, int *page_started, int force,
1146 unsigned long *nr_written)
1148 struct btrfs_root *root = BTRFS_I(inode)->root;
1149 struct btrfs_trans_handle *trans;
1150 struct extent_buffer *leaf;
1151 struct btrfs_path *path;
1152 struct btrfs_file_extent_item *fi;
1153 struct btrfs_key found_key;
1167 u64 ino = btrfs_ino(inode);
1169 path = btrfs_alloc_path();
1171 extent_clear_unlock_delalloc(inode,
1172 &BTRFS_I(inode)->io_tree,
1173 start, end, locked_page,
1174 EXTENT_CLEAR_UNLOCK_PAGE |
1175 EXTENT_CLEAR_UNLOCK |
1176 EXTENT_CLEAR_DELALLOC |
1177 EXTENT_CLEAR_DIRTY |
1178 EXTENT_SET_WRITEBACK |
1179 EXTENT_END_WRITEBACK);
1183 nolock = btrfs_is_free_space_inode(inode);
1186 trans = btrfs_join_transaction_nolock(root);
1188 trans = btrfs_join_transaction(root);
1190 if (IS_ERR(trans)) {
1191 extent_clear_unlock_delalloc(inode,
1192 &BTRFS_I(inode)->io_tree,
1193 start, end, locked_page,
1194 EXTENT_CLEAR_UNLOCK_PAGE |
1195 EXTENT_CLEAR_UNLOCK |
1196 EXTENT_CLEAR_DELALLOC |
1197 EXTENT_CLEAR_DIRTY |
1198 EXTENT_SET_WRITEBACK |
1199 EXTENT_END_WRITEBACK);
1200 btrfs_free_path(path);
1201 return PTR_ERR(trans);
1204 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1206 cow_start = (u64)-1;
1209 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1212 btrfs_abort_transaction(trans, root, ret);
1215 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1216 leaf = path->nodes[0];
1217 btrfs_item_key_to_cpu(leaf, &found_key,
1218 path->slots[0] - 1);
1219 if (found_key.objectid == ino &&
1220 found_key.type == BTRFS_EXTENT_DATA_KEY)
1225 leaf = path->nodes[0];
1226 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1227 ret = btrfs_next_leaf(root, path);
1229 btrfs_abort_transaction(trans, root, ret);
1234 leaf = path->nodes[0];
1240 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1242 if (found_key.objectid > ino ||
1243 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1244 found_key.offset > end)
1247 if (found_key.offset > cur_offset) {
1248 extent_end = found_key.offset;
1253 fi = btrfs_item_ptr(leaf, path->slots[0],
1254 struct btrfs_file_extent_item);
1255 extent_type = btrfs_file_extent_type(leaf, fi);
1257 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1258 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1259 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1260 extent_offset = btrfs_file_extent_offset(leaf, fi);
1261 extent_end = found_key.offset +
1262 btrfs_file_extent_num_bytes(leaf, fi);
1264 btrfs_file_extent_disk_num_bytes(leaf, fi);
1265 if (extent_end <= start) {
1269 if (disk_bytenr == 0)
1271 if (btrfs_file_extent_compression(leaf, fi) ||
1272 btrfs_file_extent_encryption(leaf, fi) ||
1273 btrfs_file_extent_other_encoding(leaf, fi))
1275 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1277 if (btrfs_extent_readonly(root, disk_bytenr))
1279 if (btrfs_cross_ref_exist(trans, root, ino,
1281 extent_offset, disk_bytenr))
1283 disk_bytenr += extent_offset;
1284 disk_bytenr += cur_offset - found_key.offset;
1285 num_bytes = min(end + 1, extent_end) - cur_offset;
1287 * force cow if csum exists in the range.
1288 * this ensure that csum for a given extent are
1289 * either valid or do not exist.
1291 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1294 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1295 extent_end = found_key.offset +
1296 btrfs_file_extent_inline_len(leaf, fi);
1297 extent_end = ALIGN(extent_end, root->sectorsize);
1302 if (extent_end <= start) {
1307 if (cow_start == (u64)-1)
1308 cow_start = cur_offset;
1309 cur_offset = extent_end;
1310 if (cur_offset > end)
1316 btrfs_release_path(path);
1317 if (cow_start != (u64)-1) {
1318 ret = __cow_file_range(trans, inode, root, locked_page,
1319 cow_start, found_key.offset - 1,
1320 page_started, nr_written, 1);
1322 btrfs_abort_transaction(trans, root, ret);
1325 cow_start = (u64)-1;
1328 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1329 struct extent_map *em;
1330 struct extent_map_tree *em_tree;
1331 em_tree = &BTRFS_I(inode)->extent_tree;
1332 em = alloc_extent_map();
1333 BUG_ON(!em); /* -ENOMEM */
1334 em->start = cur_offset;
1335 em->orig_start = found_key.offset - extent_offset;
1336 em->len = num_bytes;
1337 em->block_len = num_bytes;
1338 em->block_start = disk_bytenr;
1339 em->orig_block_len = disk_num_bytes;
1340 em->bdev = root->fs_info->fs_devices->latest_bdev;
1341 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1342 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1343 em->generation = -1;
1345 write_lock(&em_tree->lock);
1346 ret = add_extent_mapping(em_tree, em);
1348 list_move(&em->list,
1349 &em_tree->modified_extents);
1350 write_unlock(&em_tree->lock);
1351 if (ret != -EEXIST) {
1352 free_extent_map(em);
1355 btrfs_drop_extent_cache(inode, em->start,
1356 em->start + em->len - 1, 0);
1358 type = BTRFS_ORDERED_PREALLOC;
1360 type = BTRFS_ORDERED_NOCOW;
1363 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1364 num_bytes, num_bytes, type);
1365 BUG_ON(ret); /* -ENOMEM */
1367 if (root->root_key.objectid ==
1368 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1369 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1372 btrfs_abort_transaction(trans, root, ret);
1377 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1378 cur_offset, cur_offset + num_bytes - 1,
1379 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1380 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1381 EXTENT_SET_PRIVATE2);
1382 cur_offset = extent_end;
1383 if (cur_offset > end)
1386 btrfs_release_path(path);
1388 if (cur_offset <= end && cow_start == (u64)-1) {
1389 cow_start = cur_offset;
1393 if (cow_start != (u64)-1) {
1394 ret = __cow_file_range(trans, inode, root, locked_page,
1396 page_started, nr_written, 1);
1398 btrfs_abort_transaction(trans, root, ret);
1404 err = btrfs_end_transaction(trans, root);
1408 if (ret && cur_offset < end)
1409 extent_clear_unlock_delalloc(inode,
1410 &BTRFS_I(inode)->io_tree,
1411 cur_offset, end, locked_page,
1412 EXTENT_CLEAR_UNLOCK_PAGE |
1413 EXTENT_CLEAR_UNLOCK |
1414 EXTENT_CLEAR_DELALLOC |
1415 EXTENT_CLEAR_DIRTY |
1416 EXTENT_SET_WRITEBACK |
1417 EXTENT_END_WRITEBACK);
1419 btrfs_free_path(path);
1424 * extent_io.c call back to do delayed allocation processing
1426 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1427 u64 start, u64 end, int *page_started,
1428 unsigned long *nr_written)
1431 struct btrfs_root *root = BTRFS_I(inode)->root;
1433 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1434 ret = run_delalloc_nocow(inode, locked_page, start, end,
1435 page_started, 1, nr_written);
1436 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1437 ret = run_delalloc_nocow(inode, locked_page, start, end,
1438 page_started, 0, nr_written);
1439 } else if (!btrfs_test_opt(root, COMPRESS) &&
1440 !(BTRFS_I(inode)->force_compress) &&
1441 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1442 ret = cow_file_range(inode, locked_page, start, end,
1443 page_started, nr_written, 1);
1445 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1446 &BTRFS_I(inode)->runtime_flags);
1447 ret = cow_file_range_async(inode, locked_page, start, end,
1448 page_started, nr_written);
1453 static void btrfs_split_extent_hook(struct inode *inode,
1454 struct extent_state *orig, u64 split)
1456 /* not delalloc, ignore it */
1457 if (!(orig->state & EXTENT_DELALLOC))
1460 spin_lock(&BTRFS_I(inode)->lock);
1461 BTRFS_I(inode)->outstanding_extents++;
1462 spin_unlock(&BTRFS_I(inode)->lock);
1466 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1467 * extents so we can keep track of new extents that are just merged onto old
1468 * extents, such as when we are doing sequential writes, so we can properly
1469 * account for the metadata space we'll need.
1471 static void btrfs_merge_extent_hook(struct inode *inode,
1472 struct extent_state *new,
1473 struct extent_state *other)
1475 /* not delalloc, ignore it */
1476 if (!(other->state & EXTENT_DELALLOC))
1479 spin_lock(&BTRFS_I(inode)->lock);
1480 BTRFS_I(inode)->outstanding_extents--;
1481 spin_unlock(&BTRFS_I(inode)->lock);
1485 * extent_io.c set_bit_hook, used to track delayed allocation
1486 * bytes in this file, and to maintain the list of inodes that
1487 * have pending delalloc work to be done.
1489 static void btrfs_set_bit_hook(struct inode *inode,
1490 struct extent_state *state, int *bits)
1494 * set_bit and clear bit hooks normally require _irqsave/restore
1495 * but in this case, we are only testing for the DELALLOC
1496 * bit, which is only set or cleared with irqs on
1498 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1499 struct btrfs_root *root = BTRFS_I(inode)->root;
1500 u64 len = state->end + 1 - state->start;
1501 bool do_list = !btrfs_is_free_space_inode(inode);
1503 if (*bits & EXTENT_FIRST_DELALLOC) {
1504 *bits &= ~EXTENT_FIRST_DELALLOC;
1506 spin_lock(&BTRFS_I(inode)->lock);
1507 BTRFS_I(inode)->outstanding_extents++;
1508 spin_unlock(&BTRFS_I(inode)->lock);
1511 spin_lock(&root->fs_info->delalloc_lock);
1512 BTRFS_I(inode)->delalloc_bytes += len;
1513 root->fs_info->delalloc_bytes += len;
1514 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1515 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1516 &root->fs_info->delalloc_inodes);
1518 spin_unlock(&root->fs_info->delalloc_lock);
1523 * extent_io.c clear_bit_hook, see set_bit_hook for why
1525 static void btrfs_clear_bit_hook(struct inode *inode,
1526 struct extent_state *state, int *bits)
1529 * set_bit and clear bit hooks normally require _irqsave/restore
1530 * but in this case, we are only testing for the DELALLOC
1531 * bit, which is only set or cleared with irqs on
1533 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1534 struct btrfs_root *root = BTRFS_I(inode)->root;
1535 u64 len = state->end + 1 - state->start;
1536 bool do_list = !btrfs_is_free_space_inode(inode);
1538 if (*bits & EXTENT_FIRST_DELALLOC) {
1539 *bits &= ~EXTENT_FIRST_DELALLOC;
1540 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541 spin_lock(&BTRFS_I(inode)->lock);
1542 BTRFS_I(inode)->outstanding_extents--;
1543 spin_unlock(&BTRFS_I(inode)->lock);
1546 if (*bits & EXTENT_DO_ACCOUNTING)
1547 btrfs_delalloc_release_metadata(inode, len);
1549 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1551 btrfs_free_reserved_data_space(inode, len);
1553 spin_lock(&root->fs_info->delalloc_lock);
1554 root->fs_info->delalloc_bytes -= len;
1555 BTRFS_I(inode)->delalloc_bytes -= len;
1557 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1558 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1559 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1561 spin_unlock(&root->fs_info->delalloc_lock);
1566 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1567 * we don't create bios that span stripes or chunks
1569 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1570 size_t size, struct bio *bio,
1571 unsigned long bio_flags)
1573 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1574 u64 logical = (u64)bio->bi_sector << 9;
1579 if (bio_flags & EXTENT_BIO_COMPRESSED)
1582 length = bio->bi_size;
1583 map_length = length;
1584 ret = btrfs_map_block(root->fs_info, READ, logical,
1585 &map_length, NULL, 0);
1586 /* Will always return 0 with map_multi == NULL */
1588 if (map_length < length + size)
1594 * in order to insert checksums into the metadata in large chunks,
1595 * we wait until bio submission time. All the pages in the bio are
1596 * checksummed and sums are attached onto the ordered extent record.
1598 * At IO completion time the cums attached on the ordered extent record
1599 * are inserted into the btree
1601 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1602 struct bio *bio, int mirror_num,
1603 unsigned long bio_flags,
1606 struct btrfs_root *root = BTRFS_I(inode)->root;
1609 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1610 BUG_ON(ret); /* -ENOMEM */
1615 * in order to insert checksums into the metadata in large chunks,
1616 * we wait until bio submission time. All the pages in the bio are
1617 * checksummed and sums are attached onto the ordered extent record.
1619 * At IO completion time the cums attached on the ordered extent record
1620 * are inserted into the btree
1622 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1623 int mirror_num, unsigned long bio_flags,
1626 struct btrfs_root *root = BTRFS_I(inode)->root;
1629 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1631 bio_endio(bio, ret);
1636 * extent_io.c submission hook. This does the right thing for csum calculation
1637 * on write, or reading the csums from the tree before a read
1639 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1640 int mirror_num, unsigned long bio_flags,
1643 struct btrfs_root *root = BTRFS_I(inode)->root;
1647 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1649 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1651 if (btrfs_is_free_space_inode(inode))
1654 if (!(rw & REQ_WRITE)) {
1655 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1659 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1660 ret = btrfs_submit_compressed_read(inode, bio,
1664 } else if (!skip_sum) {
1665 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1670 } else if (async && !skip_sum) {
1671 /* csum items have already been cloned */
1672 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1674 /* we're doing a write, do the async checksumming */
1675 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1676 inode, rw, bio, mirror_num,
1677 bio_flags, bio_offset,
1678 __btrfs_submit_bio_start,
1679 __btrfs_submit_bio_done);
1681 } else if (!skip_sum) {
1682 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1688 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1692 bio_endio(bio, ret);
1697 * given a list of ordered sums record them in the inode. This happens
1698 * at IO completion time based on sums calculated at bio submission time.
1700 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1701 struct inode *inode, u64 file_offset,
1702 struct list_head *list)
1704 struct btrfs_ordered_sum *sum;
1706 list_for_each_entry(sum, list, list) {
1707 btrfs_csum_file_blocks(trans,
1708 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1713 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1714 struct extent_state **cached_state)
1716 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1717 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1718 cached_state, GFP_NOFS);
1721 /* see btrfs_writepage_start_hook for details on why this is required */
1722 struct btrfs_writepage_fixup {
1724 struct btrfs_work work;
1727 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1729 struct btrfs_writepage_fixup *fixup;
1730 struct btrfs_ordered_extent *ordered;
1731 struct extent_state *cached_state = NULL;
1733 struct inode *inode;
1738 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1742 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1743 ClearPageChecked(page);
1747 inode = page->mapping->host;
1748 page_start = page_offset(page);
1749 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1751 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1754 /* already ordered? We're done */
1755 if (PagePrivate2(page))
1758 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1760 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1761 page_end, &cached_state, GFP_NOFS);
1763 btrfs_start_ordered_extent(inode, ordered, 1);
1764 btrfs_put_ordered_extent(ordered);
1768 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1770 mapping_set_error(page->mapping, ret);
1771 end_extent_writepage(page, ret, page_start, page_end);
1772 ClearPageChecked(page);
1776 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1777 ClearPageChecked(page);
1778 set_page_dirty(page);
1780 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1781 &cached_state, GFP_NOFS);
1784 page_cache_release(page);
1789 * There are a few paths in the higher layers of the kernel that directly
1790 * set the page dirty bit without asking the filesystem if it is a
1791 * good idea. This causes problems because we want to make sure COW
1792 * properly happens and the data=ordered rules are followed.
1794 * In our case any range that doesn't have the ORDERED bit set
1795 * hasn't been properly setup for IO. We kick off an async process
1796 * to fix it up. The async helper will wait for ordered extents, set
1797 * the delalloc bit and make it safe to write the page.
1799 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1801 struct inode *inode = page->mapping->host;
1802 struct btrfs_writepage_fixup *fixup;
1803 struct btrfs_root *root = BTRFS_I(inode)->root;
1805 /* this page is properly in the ordered list */
1806 if (TestClearPagePrivate2(page))
1809 if (PageChecked(page))
1812 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1816 SetPageChecked(page);
1817 page_cache_get(page);
1818 fixup->work.func = btrfs_writepage_fixup_worker;
1820 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1824 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1825 struct inode *inode, u64 file_pos,
1826 u64 disk_bytenr, u64 disk_num_bytes,
1827 u64 num_bytes, u64 ram_bytes,
1828 u8 compression, u8 encryption,
1829 u16 other_encoding, int extent_type)
1831 struct btrfs_root *root = BTRFS_I(inode)->root;
1832 struct btrfs_file_extent_item *fi;
1833 struct btrfs_path *path;
1834 struct extent_buffer *leaf;
1835 struct btrfs_key ins;
1838 path = btrfs_alloc_path();
1842 path->leave_spinning = 1;
1845 * we may be replacing one extent in the tree with another.
1846 * The new extent is pinned in the extent map, and we don't want
1847 * to drop it from the cache until it is completely in the btree.
1849 * So, tell btrfs_drop_extents to leave this extent in the cache.
1850 * the caller is expected to unpin it and allow it to be merged
1853 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1854 file_pos + num_bytes, 0);
1858 ins.objectid = btrfs_ino(inode);
1859 ins.offset = file_pos;
1860 ins.type = BTRFS_EXTENT_DATA_KEY;
1861 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1864 leaf = path->nodes[0];
1865 fi = btrfs_item_ptr(leaf, path->slots[0],
1866 struct btrfs_file_extent_item);
1867 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1868 btrfs_set_file_extent_type(leaf, fi, extent_type);
1869 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1870 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1871 btrfs_set_file_extent_offset(leaf, fi, 0);
1872 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1873 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1874 btrfs_set_file_extent_compression(leaf, fi, compression);
1875 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1876 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1878 btrfs_mark_buffer_dirty(leaf);
1879 btrfs_release_path(path);
1881 inode_add_bytes(inode, num_bytes);
1883 ins.objectid = disk_bytenr;
1884 ins.offset = disk_num_bytes;
1885 ins.type = BTRFS_EXTENT_ITEM_KEY;
1886 ret = btrfs_alloc_reserved_file_extent(trans, root,
1887 root->root_key.objectid,
1888 btrfs_ino(inode), file_pos, &ins);
1890 btrfs_free_path(path);
1896 * helper function for btrfs_finish_ordered_io, this
1897 * just reads in some of the csum leaves to prime them into ram
1898 * before we start the transaction. It limits the amount of btree
1899 * reads required while inside the transaction.
1901 /* as ordered data IO finishes, this gets called so we can finish
1902 * an ordered extent if the range of bytes in the file it covers are
1905 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1907 struct inode *inode = ordered_extent->inode;
1908 struct btrfs_root *root = BTRFS_I(inode)->root;
1909 struct btrfs_trans_handle *trans = NULL;
1910 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1911 struct extent_state *cached_state = NULL;
1912 int compress_type = 0;
1916 nolock = btrfs_is_free_space_inode(inode);
1918 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1923 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1924 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1925 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1927 trans = btrfs_join_transaction_nolock(root);
1929 trans = btrfs_join_transaction(root);
1930 if (IS_ERR(trans)) {
1931 ret = PTR_ERR(trans);
1935 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1936 ret = btrfs_update_inode_fallback(trans, root, inode);
1937 if (ret) /* -ENOMEM or corruption */
1938 btrfs_abort_transaction(trans, root, ret);
1942 lock_extent_bits(io_tree, ordered_extent->file_offset,
1943 ordered_extent->file_offset + ordered_extent->len - 1,
1947 trans = btrfs_join_transaction_nolock(root);
1949 trans = btrfs_join_transaction(root);
1950 if (IS_ERR(trans)) {
1951 ret = PTR_ERR(trans);
1955 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1957 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1958 compress_type = ordered_extent->compress_type;
1959 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1960 BUG_ON(compress_type);
1961 ret = btrfs_mark_extent_written(trans, inode,
1962 ordered_extent->file_offset,
1963 ordered_extent->file_offset +
1964 ordered_extent->len);
1966 BUG_ON(root == root->fs_info->tree_root);
1967 ret = insert_reserved_file_extent(trans, inode,
1968 ordered_extent->file_offset,
1969 ordered_extent->start,
1970 ordered_extent->disk_len,
1971 ordered_extent->len,
1972 ordered_extent->len,
1973 compress_type, 0, 0,
1974 BTRFS_FILE_EXTENT_REG);
1976 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1977 ordered_extent->file_offset, ordered_extent->len,
1980 btrfs_abort_transaction(trans, root, ret);
1984 add_pending_csums(trans, inode, ordered_extent->file_offset,
1985 &ordered_extent->list);
1987 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1988 ret = btrfs_update_inode_fallback(trans, root, inode);
1989 if (ret) { /* -ENOMEM or corruption */
1990 btrfs_abort_transaction(trans, root, ret);
1995 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1996 ordered_extent->file_offset +
1997 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1999 if (root != root->fs_info->tree_root)
2000 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2002 btrfs_end_transaction(trans, root);
2005 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2006 ordered_extent->file_offset +
2007 ordered_extent->len - 1, NULL, GFP_NOFS);
2010 * This needs to be done to make sure anybody waiting knows we are done
2011 * updating everything for this ordered extent.
2013 btrfs_remove_ordered_extent(inode, ordered_extent);
2016 btrfs_put_ordered_extent(ordered_extent);
2017 /* once for the tree */
2018 btrfs_put_ordered_extent(ordered_extent);
2023 static void finish_ordered_fn(struct btrfs_work *work)
2025 struct btrfs_ordered_extent *ordered_extent;
2026 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2027 btrfs_finish_ordered_io(ordered_extent);
2030 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2031 struct extent_state *state, int uptodate)
2033 struct inode *inode = page->mapping->host;
2034 struct btrfs_root *root = BTRFS_I(inode)->root;
2035 struct btrfs_ordered_extent *ordered_extent = NULL;
2036 struct btrfs_workers *workers;
2038 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2040 ClearPagePrivate2(page);
2041 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2042 end - start + 1, uptodate))
2045 ordered_extent->work.func = finish_ordered_fn;
2046 ordered_extent->work.flags = 0;
2048 if (btrfs_is_free_space_inode(inode))
2049 workers = &root->fs_info->endio_freespace_worker;
2051 workers = &root->fs_info->endio_write_workers;
2052 btrfs_queue_worker(workers, &ordered_extent->work);
2058 * when reads are done, we need to check csums to verify the data is correct
2059 * if there's a match, we allow the bio to finish. If not, the code in
2060 * extent_io.c will try to find good copies for us.
2062 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2063 struct extent_state *state, int mirror)
2065 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2066 struct inode *inode = page->mapping->host;
2067 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2069 u64 private = ~(u32)0;
2071 struct btrfs_root *root = BTRFS_I(inode)->root;
2074 if (PageChecked(page)) {
2075 ClearPageChecked(page);
2079 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2082 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2083 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2084 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2089 if (state && state->start == start) {
2090 private = state->private;
2093 ret = get_state_private(io_tree, start, &private);
2095 kaddr = kmap_atomic(page);
2099 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2100 btrfs_csum_final(csum, (char *)&csum);
2101 if (csum != private)
2104 kunmap_atomic(kaddr);
2109 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2111 (unsigned long long)btrfs_ino(page->mapping->host),
2112 (unsigned long long)start, csum,
2113 (unsigned long long)private);
2114 memset(kaddr + offset, 1, end - start + 1);
2115 flush_dcache_page(page);
2116 kunmap_atomic(kaddr);
2122 struct delayed_iput {
2123 struct list_head list;
2124 struct inode *inode;
2127 /* JDM: If this is fs-wide, why can't we add a pointer to
2128 * btrfs_inode instead and avoid the allocation? */
2129 void btrfs_add_delayed_iput(struct inode *inode)
2131 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2132 struct delayed_iput *delayed;
2134 if (atomic_add_unless(&inode->i_count, -1, 1))
2137 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2138 delayed->inode = inode;
2140 spin_lock(&fs_info->delayed_iput_lock);
2141 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2142 spin_unlock(&fs_info->delayed_iput_lock);
2145 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2148 struct btrfs_fs_info *fs_info = root->fs_info;
2149 struct delayed_iput *delayed;
2152 spin_lock(&fs_info->delayed_iput_lock);
2153 empty = list_empty(&fs_info->delayed_iputs);
2154 spin_unlock(&fs_info->delayed_iput_lock);
2158 spin_lock(&fs_info->delayed_iput_lock);
2159 list_splice_init(&fs_info->delayed_iputs, &list);
2160 spin_unlock(&fs_info->delayed_iput_lock);
2162 while (!list_empty(&list)) {
2163 delayed = list_entry(list.next, struct delayed_iput, list);
2164 list_del(&delayed->list);
2165 iput(delayed->inode);
2170 enum btrfs_orphan_cleanup_state {
2171 ORPHAN_CLEANUP_STARTED = 1,
2172 ORPHAN_CLEANUP_DONE = 2,
2176 * This is called in transaction commit time. If there are no orphan
2177 * files in the subvolume, it removes orphan item and frees block_rsv
2180 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2181 struct btrfs_root *root)
2183 struct btrfs_block_rsv *block_rsv;
2186 if (atomic_read(&root->orphan_inodes) ||
2187 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2190 spin_lock(&root->orphan_lock);
2191 if (atomic_read(&root->orphan_inodes)) {
2192 spin_unlock(&root->orphan_lock);
2196 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2197 spin_unlock(&root->orphan_lock);
2201 block_rsv = root->orphan_block_rsv;
2202 root->orphan_block_rsv = NULL;
2203 spin_unlock(&root->orphan_lock);
2205 if (root->orphan_item_inserted &&
2206 btrfs_root_refs(&root->root_item) > 0) {
2207 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2208 root->root_key.objectid);
2210 root->orphan_item_inserted = 0;
2214 WARN_ON(block_rsv->size > 0);
2215 btrfs_free_block_rsv(root, block_rsv);
2220 * This creates an orphan entry for the given inode in case something goes
2221 * wrong in the middle of an unlink/truncate.
2223 * NOTE: caller of this function should reserve 5 units of metadata for
2226 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2228 struct btrfs_root *root = BTRFS_I(inode)->root;
2229 struct btrfs_block_rsv *block_rsv = NULL;
2234 if (!root->orphan_block_rsv) {
2235 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2240 spin_lock(&root->orphan_lock);
2241 if (!root->orphan_block_rsv) {
2242 root->orphan_block_rsv = block_rsv;
2243 } else if (block_rsv) {
2244 btrfs_free_block_rsv(root, block_rsv);
2248 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2249 &BTRFS_I(inode)->runtime_flags)) {
2252 * For proper ENOSPC handling, we should do orphan
2253 * cleanup when mounting. But this introduces backward
2254 * compatibility issue.
2256 if (!xchg(&root->orphan_item_inserted, 1))
2262 atomic_inc(&root->orphan_inodes);
2265 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2266 &BTRFS_I(inode)->runtime_flags))
2268 spin_unlock(&root->orphan_lock);
2270 /* grab metadata reservation from transaction handle */
2272 ret = btrfs_orphan_reserve_metadata(trans, inode);
2273 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2276 /* insert an orphan item to track this unlinked/truncated file */
2278 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2279 if (ret && ret != -EEXIST) {
2280 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2281 &BTRFS_I(inode)->runtime_flags);
2282 btrfs_abort_transaction(trans, root, ret);
2288 /* insert an orphan item to track subvolume contains orphan files */
2290 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2291 root->root_key.objectid);
2292 if (ret && ret != -EEXIST) {
2293 btrfs_abort_transaction(trans, root, ret);
2301 * We have done the truncate/delete so we can go ahead and remove the orphan
2302 * item for this particular inode.
2304 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2306 struct btrfs_root *root = BTRFS_I(inode)->root;
2307 int delete_item = 0;
2308 int release_rsv = 0;
2311 spin_lock(&root->orphan_lock);
2312 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2313 &BTRFS_I(inode)->runtime_flags))
2316 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2317 &BTRFS_I(inode)->runtime_flags))
2319 spin_unlock(&root->orphan_lock);
2321 if (trans && delete_item) {
2322 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2323 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2327 btrfs_orphan_release_metadata(inode);
2328 atomic_dec(&root->orphan_inodes);
2335 * this cleans up any orphans that may be left on the list from the last use
2338 int btrfs_orphan_cleanup(struct btrfs_root *root)
2340 struct btrfs_path *path;
2341 struct extent_buffer *leaf;
2342 struct btrfs_key key, found_key;
2343 struct btrfs_trans_handle *trans;
2344 struct inode *inode;
2345 u64 last_objectid = 0;
2346 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2348 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2351 path = btrfs_alloc_path();
2358 key.objectid = BTRFS_ORPHAN_OBJECTID;
2359 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2360 key.offset = (u64)-1;
2363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2368 * if ret == 0 means we found what we were searching for, which
2369 * is weird, but possible, so only screw with path if we didn't
2370 * find the key and see if we have stuff that matches
2374 if (path->slots[0] == 0)
2379 /* pull out the item */
2380 leaf = path->nodes[0];
2381 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2383 /* make sure the item matches what we want */
2384 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2386 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2389 /* release the path since we're done with it */
2390 btrfs_release_path(path);
2393 * this is where we are basically btrfs_lookup, without the
2394 * crossing root thing. we store the inode number in the
2395 * offset of the orphan item.
2398 if (found_key.offset == last_objectid) {
2399 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2400 "stopping orphan cleanup\n");
2405 last_objectid = found_key.offset;
2407 found_key.objectid = found_key.offset;
2408 found_key.type = BTRFS_INODE_ITEM_KEY;
2409 found_key.offset = 0;
2410 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2411 ret = PTR_RET(inode);
2412 if (ret && ret != -ESTALE)
2415 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2416 struct btrfs_root *dead_root;
2417 struct btrfs_fs_info *fs_info = root->fs_info;
2418 int is_dead_root = 0;
2421 * this is an orphan in the tree root. Currently these
2422 * could come from 2 sources:
2423 * a) a snapshot deletion in progress
2424 * b) a free space cache inode
2425 * We need to distinguish those two, as the snapshot
2426 * orphan must not get deleted.
2427 * find_dead_roots already ran before us, so if this
2428 * is a snapshot deletion, we should find the root
2429 * in the dead_roots list
2431 spin_lock(&fs_info->trans_lock);
2432 list_for_each_entry(dead_root, &fs_info->dead_roots,
2434 if (dead_root->root_key.objectid ==
2435 found_key.objectid) {
2440 spin_unlock(&fs_info->trans_lock);
2442 /* prevent this orphan from being found again */
2443 key.offset = found_key.objectid - 1;
2448 * Inode is already gone but the orphan item is still there,
2449 * kill the orphan item.
2451 if (ret == -ESTALE) {
2452 trans = btrfs_start_transaction(root, 1);
2453 if (IS_ERR(trans)) {
2454 ret = PTR_ERR(trans);
2457 printk(KERN_ERR "auto deleting %Lu\n",
2458 found_key.objectid);
2459 ret = btrfs_del_orphan_item(trans, root,
2460 found_key.objectid);
2461 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2462 btrfs_end_transaction(trans, root);
2467 * add this inode to the orphan list so btrfs_orphan_del does
2468 * the proper thing when we hit it
2470 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2471 &BTRFS_I(inode)->runtime_flags);
2473 /* if we have links, this was a truncate, lets do that */
2474 if (inode->i_nlink) {
2475 if (!S_ISREG(inode->i_mode)) {
2482 /* 1 for the orphan item deletion. */
2483 trans = btrfs_start_transaction(root, 1);
2484 if (IS_ERR(trans)) {
2485 ret = PTR_ERR(trans);
2488 ret = btrfs_orphan_add(trans, inode);
2489 btrfs_end_transaction(trans, root);
2493 ret = btrfs_truncate(inode);
2498 /* this will do delete_inode and everything for us */
2503 /* release the path since we're done with it */
2504 btrfs_release_path(path);
2506 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2508 if (root->orphan_block_rsv)
2509 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2512 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2513 trans = btrfs_join_transaction(root);
2515 btrfs_end_transaction(trans, root);
2519 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2521 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2525 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2526 btrfs_free_path(path);
2531 * very simple check to peek ahead in the leaf looking for xattrs. If we
2532 * don't find any xattrs, we know there can't be any acls.
2534 * slot is the slot the inode is in, objectid is the objectid of the inode
2536 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2537 int slot, u64 objectid)
2539 u32 nritems = btrfs_header_nritems(leaf);
2540 struct btrfs_key found_key;
2544 while (slot < nritems) {
2545 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2547 /* we found a different objectid, there must not be acls */
2548 if (found_key.objectid != objectid)
2551 /* we found an xattr, assume we've got an acl */
2552 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2556 * we found a key greater than an xattr key, there can't
2557 * be any acls later on
2559 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2566 * it goes inode, inode backrefs, xattrs, extents,
2567 * so if there are a ton of hard links to an inode there can
2568 * be a lot of backrefs. Don't waste time searching too hard,
2569 * this is just an optimization
2574 /* we hit the end of the leaf before we found an xattr or
2575 * something larger than an xattr. We have to assume the inode
2582 * read an inode from the btree into the in-memory inode
2584 static void btrfs_read_locked_inode(struct inode *inode)
2586 struct btrfs_path *path;
2587 struct extent_buffer *leaf;
2588 struct btrfs_inode_item *inode_item;
2589 struct btrfs_timespec *tspec;
2590 struct btrfs_root *root = BTRFS_I(inode)->root;
2591 struct btrfs_key location;
2595 bool filled = false;
2597 ret = btrfs_fill_inode(inode, &rdev);
2601 path = btrfs_alloc_path();
2605 path->leave_spinning = 1;
2606 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2608 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2612 leaf = path->nodes[0];
2617 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2618 struct btrfs_inode_item);
2619 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2620 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2621 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2622 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2623 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2625 tspec = btrfs_inode_atime(inode_item);
2626 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2627 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2629 tspec = btrfs_inode_mtime(inode_item);
2630 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2631 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2633 tspec = btrfs_inode_ctime(inode_item);
2634 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2635 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2637 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2638 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2639 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2642 * If we were modified in the current generation and evicted from memory
2643 * and then re-read we need to do a full sync since we don't have any
2644 * idea about which extents were modified before we were evicted from
2647 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2648 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2649 &BTRFS_I(inode)->runtime_flags);
2651 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2652 inode->i_generation = BTRFS_I(inode)->generation;
2654 rdev = btrfs_inode_rdev(leaf, inode_item);
2656 BTRFS_I(inode)->index_cnt = (u64)-1;
2657 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2660 * try to precache a NULL acl entry for files that don't have
2661 * any xattrs or acls
2663 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2666 cache_no_acl(inode);
2668 btrfs_free_path(path);
2670 switch (inode->i_mode & S_IFMT) {
2672 inode->i_mapping->a_ops = &btrfs_aops;
2673 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2674 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2675 inode->i_fop = &btrfs_file_operations;
2676 inode->i_op = &btrfs_file_inode_operations;
2679 inode->i_fop = &btrfs_dir_file_operations;
2680 if (root == root->fs_info->tree_root)
2681 inode->i_op = &btrfs_dir_ro_inode_operations;
2683 inode->i_op = &btrfs_dir_inode_operations;
2686 inode->i_op = &btrfs_symlink_inode_operations;
2687 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2688 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2691 inode->i_op = &btrfs_special_inode_operations;
2692 init_special_inode(inode, inode->i_mode, rdev);
2696 btrfs_update_iflags(inode);
2700 btrfs_free_path(path);
2701 make_bad_inode(inode);
2705 * given a leaf and an inode, copy the inode fields into the leaf
2707 static void fill_inode_item(struct btrfs_trans_handle *trans,
2708 struct extent_buffer *leaf,
2709 struct btrfs_inode_item *item,
2710 struct inode *inode)
2712 btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2713 btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2714 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2715 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2716 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2718 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2719 inode->i_atime.tv_sec);
2720 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2721 inode->i_atime.tv_nsec);
2723 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2724 inode->i_mtime.tv_sec);
2725 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2726 inode->i_mtime.tv_nsec);
2728 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2729 inode->i_ctime.tv_sec);
2730 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2731 inode->i_ctime.tv_nsec);
2733 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2734 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2735 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2736 btrfs_set_inode_transid(leaf, item, trans->transid);
2737 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2738 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2739 btrfs_set_inode_block_group(leaf, item, 0);
2743 * copy everything in the in-memory inode into the btree.
2745 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2746 struct btrfs_root *root, struct inode *inode)
2748 struct btrfs_inode_item *inode_item;
2749 struct btrfs_path *path;
2750 struct extent_buffer *leaf;
2753 path = btrfs_alloc_path();
2757 path->leave_spinning = 1;
2758 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2766 btrfs_unlock_up_safe(path, 1);
2767 leaf = path->nodes[0];
2768 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2769 struct btrfs_inode_item);
2771 fill_inode_item(trans, leaf, inode_item, inode);
2772 btrfs_mark_buffer_dirty(leaf);
2773 btrfs_set_inode_last_trans(trans, inode);
2776 btrfs_free_path(path);
2781 * copy everything in the in-memory inode into the btree.
2783 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2784 struct btrfs_root *root, struct inode *inode)
2789 * If the inode is a free space inode, we can deadlock during commit
2790 * if we put it into the delayed code.
2792 * The data relocation inode should also be directly updated
2795 if (!btrfs_is_free_space_inode(inode)
2796 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2797 btrfs_update_root_times(trans, root);
2799 ret = btrfs_delayed_update_inode(trans, root, inode);
2801 btrfs_set_inode_last_trans(trans, inode);
2805 return btrfs_update_inode_item(trans, root, inode);
2808 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2809 struct btrfs_root *root,
2810 struct inode *inode)
2814 ret = btrfs_update_inode(trans, root, inode);
2816 return btrfs_update_inode_item(trans, root, inode);
2821 * unlink helper that gets used here in inode.c and in the tree logging
2822 * recovery code. It remove a link in a directory with a given name, and
2823 * also drops the back refs in the inode to the directory
2825 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2826 struct btrfs_root *root,
2827 struct inode *dir, struct inode *inode,
2828 const char *name, int name_len)
2830 struct btrfs_path *path;
2832 struct extent_buffer *leaf;
2833 struct btrfs_dir_item *di;
2834 struct btrfs_key key;
2836 u64 ino = btrfs_ino(inode);
2837 u64 dir_ino = btrfs_ino(dir);
2839 path = btrfs_alloc_path();
2845 path->leave_spinning = 1;
2846 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2847 name, name_len, -1);
2856 leaf = path->nodes[0];
2857 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2858 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2861 btrfs_release_path(path);
2863 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2866 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2867 "inode %llu parent %llu\n", name_len, name,
2868 (unsigned long long)ino, (unsigned long long)dir_ino);
2869 btrfs_abort_transaction(trans, root, ret);
2873 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2875 btrfs_abort_transaction(trans, root, ret);
2879 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2881 if (ret != 0 && ret != -ENOENT) {
2882 btrfs_abort_transaction(trans, root, ret);
2886 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2891 btrfs_free_path(path);
2895 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2896 inode_inc_iversion(inode);
2897 inode_inc_iversion(dir);
2898 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2899 ret = btrfs_update_inode(trans, root, dir);
2904 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2905 struct btrfs_root *root,
2906 struct inode *dir, struct inode *inode,
2907 const char *name, int name_len)
2910 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2912 btrfs_drop_nlink(inode);
2913 ret = btrfs_update_inode(trans, root, inode);
2919 /* helper to check if there is any shared block in the path */
2920 static int check_path_shared(struct btrfs_root *root,
2921 struct btrfs_path *path)
2923 struct extent_buffer *eb;
2927 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2930 if (!path->nodes[level])
2932 eb = path->nodes[level];
2933 if (!btrfs_block_can_be_shared(root, eb))
2935 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2944 * helper to start transaction for unlink and rmdir.
2946 * unlink and rmdir are special in btrfs, they do not always free space.
2947 * so in enospc case, we should make sure they will free space before
2948 * allowing them to use the global metadata reservation.
2950 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2951 struct dentry *dentry)
2953 struct btrfs_trans_handle *trans;
2954 struct btrfs_root *root = BTRFS_I(dir)->root;
2955 struct btrfs_path *path;
2956 struct btrfs_dir_item *di;
2957 struct inode *inode = dentry->d_inode;
2962 u64 ino = btrfs_ino(inode);
2963 u64 dir_ino = btrfs_ino(dir);
2966 * 1 for the possible orphan item
2967 * 1 for the dir item
2968 * 1 for the dir index
2969 * 1 for the inode ref
2970 * 1 for the inode ref in the tree log
2971 * 2 for the dir entries in the log
2974 trans = btrfs_start_transaction(root, 8);
2975 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2978 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2979 return ERR_PTR(-ENOSPC);
2981 /* check if there is someone else holds reference */
2982 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2983 return ERR_PTR(-ENOSPC);
2985 if (atomic_read(&inode->i_count) > 2)
2986 return ERR_PTR(-ENOSPC);
2988 if (xchg(&root->fs_info->enospc_unlink, 1))
2989 return ERR_PTR(-ENOSPC);
2991 path = btrfs_alloc_path();
2993 root->fs_info->enospc_unlink = 0;
2994 return ERR_PTR(-ENOMEM);
2997 /* 1 for the orphan item */
2998 trans = btrfs_start_transaction(root, 1);
2999 if (IS_ERR(trans)) {
3000 btrfs_free_path(path);
3001 root->fs_info->enospc_unlink = 0;
3005 path->skip_locking = 1;
3006 path->search_commit_root = 1;
3008 ret = btrfs_lookup_inode(trans, root, path,
3009 &BTRFS_I(dir)->location, 0);
3015 if (check_path_shared(root, path))
3020 btrfs_release_path(path);
3022 ret = btrfs_lookup_inode(trans, root, path,
3023 &BTRFS_I(inode)->location, 0);
3029 if (check_path_shared(root, path))
3034 btrfs_release_path(path);
3036 if (ret == 0 && S_ISREG(inode->i_mode)) {
3037 ret = btrfs_lookup_file_extent(trans, root, path,
3043 BUG_ON(ret == 0); /* Corruption */
3044 if (check_path_shared(root, path))
3046 btrfs_release_path(path);
3054 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3055 dentry->d_name.name, dentry->d_name.len, 0);
3061 if (check_path_shared(root, path))
3067 btrfs_release_path(path);
3069 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3070 dentry->d_name.len, ino, dir_ino, 0,
3077 if (check_path_shared(root, path))
3080 btrfs_release_path(path);
3083 * This is a commit root search, if we can lookup inode item and other
3084 * relative items in the commit root, it means the transaction of
3085 * dir/file creation has been committed, and the dir index item that we
3086 * delay to insert has also been inserted into the commit root. So
3087 * we needn't worry about the delayed insertion of the dir index item
3090 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3091 dentry->d_name.name, dentry->d_name.len, 0);
3096 BUG_ON(ret == -ENOENT);
3097 if (check_path_shared(root, path))
3102 btrfs_free_path(path);
3103 /* Migrate the orphan reservation over */
3105 err = btrfs_block_rsv_migrate(trans->block_rsv,
3106 &root->fs_info->global_block_rsv,
3107 trans->bytes_reserved);
3110 btrfs_end_transaction(trans, root);
3111 root->fs_info->enospc_unlink = 0;
3112 return ERR_PTR(err);
3115 trans->block_rsv = &root->fs_info->global_block_rsv;
3119 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3120 struct btrfs_root *root)
3122 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3123 btrfs_block_rsv_release(root, trans->block_rsv,
3124 trans->bytes_reserved);
3125 trans->block_rsv = &root->fs_info->trans_block_rsv;
3126 BUG_ON(!root->fs_info->enospc_unlink);
3127 root->fs_info->enospc_unlink = 0;
3129 btrfs_end_transaction(trans, root);
3132 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3134 struct btrfs_root *root = BTRFS_I(dir)->root;
3135 struct btrfs_trans_handle *trans;
3136 struct inode *inode = dentry->d_inode;
3139 trans = __unlink_start_trans(dir, dentry);
3141 return PTR_ERR(trans);
3143 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3145 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3146 dentry->d_name.name, dentry->d_name.len);
3150 if (inode->i_nlink == 0) {
3151 ret = btrfs_orphan_add(trans, inode);
3157 __unlink_end_trans(trans, root);
3158 btrfs_btree_balance_dirty(root);
3162 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3163 struct btrfs_root *root,
3164 struct inode *dir, u64 objectid,
3165 const char *name, int name_len)
3167 struct btrfs_path *path;
3168 struct extent_buffer *leaf;
3169 struct btrfs_dir_item *di;
3170 struct btrfs_key key;
3173 u64 dir_ino = btrfs_ino(dir);
3175 path = btrfs_alloc_path();
3179 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3180 name, name_len, -1);
3181 if (IS_ERR_OR_NULL(di)) {
3189 leaf = path->nodes[0];
3190 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3191 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3192 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3194 btrfs_abort_transaction(trans, root, ret);
3197 btrfs_release_path(path);
3199 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3200 objectid, root->root_key.objectid,
3201 dir_ino, &index, name, name_len);
3203 if (ret != -ENOENT) {
3204 btrfs_abort_transaction(trans, root, ret);
3207 di = btrfs_search_dir_index_item(root, path, dir_ino,
3209 if (IS_ERR_OR_NULL(di)) {
3214 btrfs_abort_transaction(trans, root, ret);
3218 leaf = path->nodes[0];
3219 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3220 btrfs_release_path(path);
3223 btrfs_release_path(path);
3225 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3227 btrfs_abort_transaction(trans, root, ret);
3231 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3232 inode_inc_iversion(dir);
3233 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3234 ret = btrfs_update_inode_fallback(trans, root, dir);
3236 btrfs_abort_transaction(trans, root, ret);
3238 btrfs_free_path(path);
3242 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3244 struct inode *inode = dentry->d_inode;
3246 struct btrfs_root *root = BTRFS_I(dir)->root;
3247 struct btrfs_trans_handle *trans;
3249 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3251 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3254 trans = __unlink_start_trans(dir, dentry);
3256 return PTR_ERR(trans);
3258 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3259 err = btrfs_unlink_subvol(trans, root, dir,
3260 BTRFS_I(inode)->location.objectid,
3261 dentry->d_name.name,
3262 dentry->d_name.len);
3266 err = btrfs_orphan_add(trans, inode);
3270 /* now the directory is empty */
3271 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3272 dentry->d_name.name, dentry->d_name.len);
3274 btrfs_i_size_write(inode, 0);
3276 __unlink_end_trans(trans, root);
3277 btrfs_btree_balance_dirty(root);
3283 * this can truncate away extent items, csum items and directory items.
3284 * It starts at a high offset and removes keys until it can't find
3285 * any higher than new_size
3287 * csum items that cross the new i_size are truncated to the new size
3290 * min_type is the minimum key type to truncate down to. If set to 0, this
3291 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3293 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3294 struct btrfs_root *root,
3295 struct inode *inode,
3296 u64 new_size, u32 min_type)
3298 struct btrfs_path *path;
3299 struct extent_buffer *leaf;
3300 struct btrfs_file_extent_item *fi;
3301 struct btrfs_key key;
3302 struct btrfs_key found_key;
3303 u64 extent_start = 0;
3304 u64 extent_num_bytes = 0;
3305 u64 extent_offset = 0;
3307 u64 mask = root->sectorsize - 1;
3308 u32 found_type = (u8)-1;
3311 int pending_del_nr = 0;
3312 int pending_del_slot = 0;
3313 int extent_type = -1;
3316 u64 ino = btrfs_ino(inode);
3318 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3320 path = btrfs_alloc_path();
3326 * We want to drop from the next block forward in case this new size is
3327 * not block aligned since we will be keeping the last block of the
3328 * extent just the way it is.
3330 if (root->ref_cows || root == root->fs_info->tree_root)
3331 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3334 * This function is also used to drop the items in the log tree before
3335 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3336 * it is used to drop the loged items. So we shouldn't kill the delayed
3339 if (min_type == 0 && root == BTRFS_I(inode)->root)
3340 btrfs_kill_delayed_inode_items(inode);
3343 key.offset = (u64)-1;
3347 path->leave_spinning = 1;
3348 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3355 /* there are no items in the tree for us to truncate, we're
3358 if (path->slots[0] == 0)
3365 leaf = path->nodes[0];
3366 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3367 found_type = btrfs_key_type(&found_key);
3369 if (found_key.objectid != ino)
3372 if (found_type < min_type)
3375 item_end = found_key.offset;
3376 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3377 fi = btrfs_item_ptr(leaf, path->slots[0],
3378 struct btrfs_file_extent_item);
3379 extent_type = btrfs_file_extent_type(leaf, fi);
3380 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3382 btrfs_file_extent_num_bytes(leaf, fi);
3383 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3384 item_end += btrfs_file_extent_inline_len(leaf,
3389 if (found_type > min_type) {
3392 if (item_end < new_size)
3394 if (found_key.offset >= new_size)
3400 /* FIXME, shrink the extent if the ref count is only 1 */
3401 if (found_type != BTRFS_EXTENT_DATA_KEY)
3404 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3406 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3408 u64 orig_num_bytes =
3409 btrfs_file_extent_num_bytes(leaf, fi);
3410 extent_num_bytes = new_size -
3411 found_key.offset + root->sectorsize - 1;
3412 extent_num_bytes = extent_num_bytes &
3413 ~((u64)root->sectorsize - 1);
3414 btrfs_set_file_extent_num_bytes(leaf, fi,
3416 num_dec = (orig_num_bytes -
3418 if (root->ref_cows && extent_start != 0)
3419 inode_sub_bytes(inode, num_dec);
3420 btrfs_mark_buffer_dirty(leaf);
3423 btrfs_file_extent_disk_num_bytes(leaf,
3425 extent_offset = found_key.offset -
3426 btrfs_file_extent_offset(leaf, fi);
3428 /* FIXME blocksize != 4096 */
3429 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3430 if (extent_start != 0) {
3433 inode_sub_bytes(inode, num_dec);
3436 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3438 * we can't truncate inline items that have had
3442 btrfs_file_extent_compression(leaf, fi) == 0 &&
3443 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3444 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3445 u32 size = new_size - found_key.offset;
3447 if (root->ref_cows) {
3448 inode_sub_bytes(inode, item_end + 1 -
3452 btrfs_file_extent_calc_inline_size(size);
3453 btrfs_truncate_item(trans, root, path,
3455 } else if (root->ref_cows) {
3456 inode_sub_bytes(inode, item_end + 1 -
3462 if (!pending_del_nr) {
3463 /* no pending yet, add ourselves */
3464 pending_del_slot = path->slots[0];
3466 } else if (pending_del_nr &&
3467 path->slots[0] + 1 == pending_del_slot) {
3468 /* hop on the pending chunk */
3470 pending_del_slot = path->slots[0];
3477 if (found_extent && (root->ref_cows ||
3478 root == root->fs_info->tree_root)) {
3479 btrfs_set_path_blocking(path);
3480 ret = btrfs_free_extent(trans, root, extent_start,
3481 extent_num_bytes, 0,
3482 btrfs_header_owner(leaf),
3483 ino, extent_offset, 0);
3487 if (found_type == BTRFS_INODE_ITEM_KEY)
3490 if (path->slots[0] == 0 ||
3491 path->slots[0] != pending_del_slot) {
3492 if (pending_del_nr) {
3493 ret = btrfs_del_items(trans, root, path,
3497 btrfs_abort_transaction(trans,
3503 btrfs_release_path(path);
3510 if (pending_del_nr) {
3511 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3514 btrfs_abort_transaction(trans, root, ret);
3517 btrfs_free_path(path);
3522 * btrfs_truncate_page - read, zero a chunk and write a page
3523 * @inode - inode that we're zeroing
3524 * @from - the offset to start zeroing
3525 * @len - the length to zero, 0 to zero the entire range respective to the
3527 * @front - zero up to the offset instead of from the offset on
3529 * This will find the page for the "from" offset and cow the page and zero the
3530 * part we want to zero. This is used with truncate and hole punching.
3532 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3535 struct address_space *mapping = inode->i_mapping;
3536 struct btrfs_root *root = BTRFS_I(inode)->root;
3537 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3538 struct btrfs_ordered_extent *ordered;
3539 struct extent_state *cached_state = NULL;
3541 u32 blocksize = root->sectorsize;
3542 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3543 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3545 gfp_t mask = btrfs_alloc_write_mask(mapping);
3550 if ((offset & (blocksize - 1)) == 0 &&
3551 (!len || ((len & (blocksize - 1)) == 0)))
3553 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3558 page = find_or_create_page(mapping, index, mask);
3560 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3565 page_start = page_offset(page);
3566 page_end = page_start + PAGE_CACHE_SIZE - 1;
3568 if (!PageUptodate(page)) {
3569 ret = btrfs_readpage(NULL, page);
3571 if (page->mapping != mapping) {
3573 page_cache_release(page);
3576 if (!PageUptodate(page)) {
3581 wait_on_page_writeback(page);
3583 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3584 set_page_extent_mapped(page);
3586 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3588 unlock_extent_cached(io_tree, page_start, page_end,
3589 &cached_state, GFP_NOFS);
3591 page_cache_release(page);
3592 btrfs_start_ordered_extent(inode, ordered, 1);
3593 btrfs_put_ordered_extent(ordered);
3597 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3598 EXTENT_DIRTY | EXTENT_DELALLOC |
3599 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3600 0, 0, &cached_state, GFP_NOFS);
3602 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3605 unlock_extent_cached(io_tree, page_start, page_end,
3606 &cached_state, GFP_NOFS);
3610 if (offset != PAGE_CACHE_SIZE) {
3612 len = PAGE_CACHE_SIZE - offset;
3615 memset(kaddr, 0, offset);
3617 memset(kaddr + offset, 0, len);
3618 flush_dcache_page(page);
3621 ClearPageChecked(page);
3622 set_page_dirty(page);
3623 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3628 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3630 page_cache_release(page);
3636 * This function puts in dummy file extents for the area we're creating a hole
3637 * for. So if we are truncating this file to a larger size we need to insert
3638 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3639 * the range between oldsize and size
3641 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3643 struct btrfs_trans_handle *trans;
3644 struct btrfs_root *root = BTRFS_I(inode)->root;
3645 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3646 struct extent_map *em = NULL;
3647 struct extent_state *cached_state = NULL;
3648 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3649 u64 mask = root->sectorsize - 1;
3650 u64 hole_start = (oldsize + mask) & ~mask;
3651 u64 block_end = (size + mask) & ~mask;
3657 if (size <= hole_start)
3661 struct btrfs_ordered_extent *ordered;
3662 btrfs_wait_ordered_range(inode, hole_start,
3663 block_end - hole_start);
3664 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3666 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3669 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3670 &cached_state, GFP_NOFS);
3671 btrfs_put_ordered_extent(ordered);
3674 cur_offset = hole_start;
3676 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3677 block_end - cur_offset, 0);
3682 last_byte = min(extent_map_end(em), block_end);
3683 last_byte = (last_byte + mask) & ~mask;
3684 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3685 struct extent_map *hole_em;
3686 hole_size = last_byte - cur_offset;
3688 trans = btrfs_start_transaction(root, 3);
3689 if (IS_ERR(trans)) {
3690 err = PTR_ERR(trans);
3694 err = btrfs_drop_extents(trans, root, inode,
3696 cur_offset + hole_size, 1);
3698 btrfs_abort_transaction(trans, root, err);
3699 btrfs_end_transaction(trans, root);
3703 err = btrfs_insert_file_extent(trans, root,
3704 btrfs_ino(inode), cur_offset, 0,
3705 0, hole_size, 0, hole_size,
3708 btrfs_abort_transaction(trans, root, err);
3709 btrfs_end_transaction(trans, root);
3713 btrfs_drop_extent_cache(inode, cur_offset,
3714 cur_offset + hole_size - 1, 0);
3715 hole_em = alloc_extent_map();
3717 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3718 &BTRFS_I(inode)->runtime_flags);
3721 hole_em->start = cur_offset;
3722 hole_em->len = hole_size;
3723 hole_em->orig_start = cur_offset;
3725 hole_em->block_start = EXTENT_MAP_HOLE;
3726 hole_em->block_len = 0;
3727 hole_em->orig_block_len = 0;
3728 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3729 hole_em->compress_type = BTRFS_COMPRESS_NONE;
3730 hole_em->generation = trans->transid;
3733 write_lock(&em_tree->lock);
3734 err = add_extent_mapping(em_tree, hole_em);
3736 list_move(&hole_em->list,
3737 &em_tree->modified_extents);
3738 write_unlock(&em_tree->lock);
3741 btrfs_drop_extent_cache(inode, cur_offset,
3745 free_extent_map(hole_em);
3747 btrfs_update_inode(trans, root, inode);
3748 btrfs_end_transaction(trans, root);
3750 free_extent_map(em);
3752 cur_offset = last_byte;
3753 if (cur_offset >= block_end)
3757 free_extent_map(em);
3758 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3763 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3765 struct btrfs_root *root = BTRFS_I(inode)->root;
3766 struct btrfs_trans_handle *trans;
3767 loff_t oldsize = i_size_read(inode);
3770 if (newsize == oldsize)
3773 if (newsize > oldsize) {
3774 truncate_pagecache(inode, oldsize, newsize);
3775 ret = btrfs_cont_expand(inode, oldsize, newsize);
3779 trans = btrfs_start_transaction(root, 1);
3781 return PTR_ERR(trans);
3783 i_size_write(inode, newsize);
3784 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3785 ret = btrfs_update_inode(trans, root, inode);
3786 btrfs_end_transaction(trans, root);
3790 * We're truncating a file that used to have good data down to
3791 * zero. Make sure it gets into the ordered flush list so that
3792 * any new writes get down to disk quickly.
3795 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3796 &BTRFS_I(inode)->runtime_flags);
3799 * 1 for the orphan item we're going to add
3800 * 1 for the orphan item deletion.
3802 trans = btrfs_start_transaction(root, 2);
3804 return PTR_ERR(trans);
3807 * We need to do this in case we fail at _any_ point during the
3808 * actual truncate. Once we do the truncate_setsize we could
3809 * invalidate pages which forces any outstanding ordered io to
3810 * be instantly completed which will give us extents that need
3811 * to be truncated. If we fail to get an orphan inode down we
3812 * could have left over extents that were never meant to live,
3813 * so we need to garuntee from this point on that everything
3814 * will be consistent.
3816 ret = btrfs_orphan_add(trans, inode);
3817 btrfs_end_transaction(trans, root);
3821 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3822 truncate_setsize(inode, newsize);
3823 ret = btrfs_truncate(inode);
3824 if (ret && inode->i_nlink)
3825 btrfs_orphan_del(NULL, inode);
3831 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3833 struct inode *inode = dentry->d_inode;
3834 struct btrfs_root *root = BTRFS_I(inode)->root;
3837 if (btrfs_root_readonly(root))
3840 err = inode_change_ok(inode, attr);
3844 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3845 err = btrfs_setsize(inode, attr->ia_size);
3850 if (attr->ia_valid) {
3851 setattr_copy(inode, attr);
3852 inode_inc_iversion(inode);
3853 err = btrfs_dirty_inode(inode);
3855 if (!err && attr->ia_valid & ATTR_MODE)
3856 err = btrfs_acl_chmod(inode);
3862 void btrfs_evict_inode(struct inode *inode)
3864 struct btrfs_trans_handle *trans;
3865 struct btrfs_root *root = BTRFS_I(inode)->root;
3866 struct btrfs_block_rsv *rsv, *global_rsv;
3867 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3870 trace_btrfs_inode_evict(inode);
3872 truncate_inode_pages(&inode->i_data, 0);
3873 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3874 btrfs_is_free_space_inode(inode)))
3877 if (is_bad_inode(inode)) {
3878 btrfs_orphan_del(NULL, inode);
3881 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3882 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3884 if (root->fs_info->log_root_recovering) {
3885 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3886 &BTRFS_I(inode)->runtime_flags));
3890 if (inode->i_nlink > 0) {
3891 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3895 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3897 btrfs_orphan_del(NULL, inode);
3900 rsv->size = min_size;
3902 global_rsv = &root->fs_info->global_block_rsv;
3904 btrfs_i_size_write(inode, 0);
3907 * This is a bit simpler than btrfs_truncate since we've already
3908 * reserved our space for our orphan item in the unlink, so we just
3909 * need to reserve some slack space in case we add bytes and update
3910 * inode item when doing the truncate.
3913 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3914 BTRFS_RESERVE_FLUSH_LIMIT);
3917 * Try and steal from the global reserve since we will
3918 * likely not use this space anyway, we want to try as
3919 * hard as possible to get this to work.
3922 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3925 printk(KERN_WARNING "Could not get space for a "
3926 "delete, will truncate on mount %d\n", ret);
3927 btrfs_orphan_del(NULL, inode);
3928 btrfs_free_block_rsv(root, rsv);
3932 trans = btrfs_start_transaction_lflush(root, 1);
3933 if (IS_ERR(trans)) {
3934 btrfs_orphan_del(NULL, inode);
3935 btrfs_free_block_rsv(root, rsv);
3939 trans->block_rsv = rsv;
3941 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3945 trans->block_rsv = &root->fs_info->trans_block_rsv;
3946 ret = btrfs_update_inode(trans, root, inode);
3949 btrfs_end_transaction(trans, root);
3951 btrfs_btree_balance_dirty(root);
3954 btrfs_free_block_rsv(root, rsv);
3957 trans->block_rsv = root->orphan_block_rsv;
3958 ret = btrfs_orphan_del(trans, inode);
3962 trans->block_rsv = &root->fs_info->trans_block_rsv;
3963 if (!(root == root->fs_info->tree_root ||
3964 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3965 btrfs_return_ino(root, btrfs_ino(inode));
3967 btrfs_end_transaction(trans, root);
3968 btrfs_btree_balance_dirty(root);
3975 * this returns the key found in the dir entry in the location pointer.
3976 * If no dir entries were found, location->objectid is 0.
3978 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3979 struct btrfs_key *location)
3981 const char *name = dentry->d_name.name;
3982 int namelen = dentry->d_name.len;
3983 struct btrfs_dir_item *di;
3984 struct btrfs_path *path;
3985 struct btrfs_root *root = BTRFS_I(dir)->root;
3988 path = btrfs_alloc_path();
3992 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3997 if (IS_ERR_OR_NULL(di))
4000 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4002 btrfs_free_path(path);
4005 location->objectid = 0;
4010 * when we hit a tree root in a directory, the btrfs part of the inode
4011 * needs to be changed to reflect the root directory of the tree root. This
4012 * is kind of like crossing a mount point.
4014 static int fixup_tree_root_location(struct btrfs_root *root,
4016 struct dentry *dentry,
4017 struct btrfs_key *location,
4018 struct btrfs_root **sub_root)
4020 struct btrfs_path *path;
4021 struct btrfs_root *new_root;
4022 struct btrfs_root_ref *ref;
4023 struct extent_buffer *leaf;
4027 path = btrfs_alloc_path();
4034 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4035 BTRFS_I(dir)->root->root_key.objectid,
4036 location->objectid);
4043 leaf = path->nodes[0];
4044 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4045 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4046 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4049 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4050 (unsigned long)(ref + 1),
4051 dentry->d_name.len);
4055 btrfs_release_path(path);
4057 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4058 if (IS_ERR(new_root)) {
4059 err = PTR_ERR(new_root);
4063 if (btrfs_root_refs(&new_root->root_item) == 0) {
4068 *sub_root = new_root;
4069 location->objectid = btrfs_root_dirid(&new_root->root_item);
4070 location->type = BTRFS_INODE_ITEM_KEY;
4071 location->offset = 0;
4074 btrfs_free_path(path);
4078 static void inode_tree_add(struct inode *inode)
4080 struct btrfs_root *root = BTRFS_I(inode)->root;
4081 struct btrfs_inode *entry;
4083 struct rb_node *parent;
4084 u64 ino = btrfs_ino(inode);
4086 p = &root->inode_tree.rb_node;
4089 if (inode_unhashed(inode))
4092 spin_lock(&root->inode_lock);
4095 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4097 if (ino < btrfs_ino(&entry->vfs_inode))
4098 p = &parent->rb_left;
4099 else if (ino > btrfs_ino(&entry->vfs_inode))
4100 p = &parent->rb_right;
4102 WARN_ON(!(entry->vfs_inode.i_state &
4103 (I_WILL_FREE | I_FREEING)));
4104 rb_erase(parent, &root->inode_tree);
4105 RB_CLEAR_NODE(parent);
4106 spin_unlock(&root->inode_lock);
4110 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4111 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4112 spin_unlock(&root->inode_lock);
4115 static void inode_tree_del(struct inode *inode)
4117 struct btrfs_root *root = BTRFS_I(inode)->root;
4120 spin_lock(&root->inode_lock);
4121 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4122 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4123 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4124 empty = RB_EMPTY_ROOT(&root->inode_tree);
4126 spin_unlock(&root->inode_lock);
4129 * Free space cache has inodes in the tree root, but the tree root has a
4130 * root_refs of 0, so this could end up dropping the tree root as a
4131 * snapshot, so we need the extra !root->fs_info->tree_root check to
4132 * make sure we don't drop it.
4134 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4135 root != root->fs_info->tree_root) {
4136 synchronize_srcu(&root->fs_info->subvol_srcu);
4137 spin_lock(&root->inode_lock);
4138 empty = RB_EMPTY_ROOT(&root->inode_tree);
4139 spin_unlock(&root->inode_lock);
4141 btrfs_add_dead_root(root);
4145 void btrfs_invalidate_inodes(struct btrfs_root *root)
4147 struct rb_node *node;
4148 struct rb_node *prev;
4149 struct btrfs_inode *entry;
4150 struct inode *inode;
4153 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4155 spin_lock(&root->inode_lock);
4157 node = root->inode_tree.rb_node;
4161 entry = rb_entry(node, struct btrfs_inode, rb_node);
4163 if (objectid < btrfs_ino(&entry->vfs_inode))
4164 node = node->rb_left;
4165 else if (objectid > btrfs_ino(&entry->vfs_inode))
4166 node = node->rb_right;
4172 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4173 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4177 prev = rb_next(prev);
4181 entry = rb_entry(node, struct btrfs_inode, rb_node);
4182 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4183 inode = igrab(&entry->vfs_inode);
4185 spin_unlock(&root->inode_lock);
4186 if (atomic_read(&inode->i_count) > 1)
4187 d_prune_aliases(inode);
4189 * btrfs_drop_inode will have it removed from
4190 * the inode cache when its usage count
4195 spin_lock(&root->inode_lock);
4199 if (cond_resched_lock(&root->inode_lock))
4202 node = rb_next(node);
4204 spin_unlock(&root->inode_lock);
4207 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4209 struct btrfs_iget_args *args = p;
4210 inode->i_ino = args->ino;
4211 BTRFS_I(inode)->root = args->root;
4215 static int btrfs_find_actor(struct inode *inode, void *opaque)
4217 struct btrfs_iget_args *args = opaque;
4218 return args->ino == btrfs_ino(inode) &&
4219 args->root == BTRFS_I(inode)->root;
4222 static struct inode *btrfs_iget_locked(struct super_block *s,
4224 struct btrfs_root *root)
4226 struct inode *inode;
4227 struct btrfs_iget_args args;
4228 args.ino = objectid;
4231 inode = iget5_locked(s, objectid, btrfs_find_actor,
4232 btrfs_init_locked_inode,
4237 /* Get an inode object given its location and corresponding root.
4238 * Returns in *is_new if the inode was read from disk
4240 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4241 struct btrfs_root *root, int *new)
4243 struct inode *inode;
4245 inode = btrfs_iget_locked(s, location->objectid, root);
4247 return ERR_PTR(-ENOMEM);
4249 if (inode->i_state & I_NEW) {
4250 BTRFS_I(inode)->root = root;
4251 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4252 btrfs_read_locked_inode(inode);
4253 if (!is_bad_inode(inode)) {
4254 inode_tree_add(inode);
4255 unlock_new_inode(inode);
4259 unlock_new_inode(inode);
4261 inode = ERR_PTR(-ESTALE);
4268 static struct inode *new_simple_dir(struct super_block *s,
4269 struct btrfs_key *key,
4270 struct btrfs_root *root)
4272 struct inode *inode = new_inode(s);
4275 return ERR_PTR(-ENOMEM);
4277 BTRFS_I(inode)->root = root;
4278 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4279 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4281 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4282 inode->i_op = &btrfs_dir_ro_inode_operations;
4283 inode->i_fop = &simple_dir_operations;
4284 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4285 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4290 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4292 struct inode *inode;
4293 struct btrfs_root *root = BTRFS_I(dir)->root;
4294 struct btrfs_root *sub_root = root;
4295 struct btrfs_key location;
4299 if (dentry->d_name.len > BTRFS_NAME_LEN)
4300 return ERR_PTR(-ENAMETOOLONG);
4302 if (unlikely(d_need_lookup(dentry))) {
4303 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4304 kfree(dentry->d_fsdata);
4305 dentry->d_fsdata = NULL;
4306 /* This thing is hashed, drop it for now */
4309 ret = btrfs_inode_by_name(dir, dentry, &location);
4313 return ERR_PTR(ret);
4315 if (location.objectid == 0)
4318 if (location.type == BTRFS_INODE_ITEM_KEY) {
4319 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4323 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4325 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4326 ret = fixup_tree_root_location(root, dir, dentry,
4327 &location, &sub_root);
4330 inode = ERR_PTR(ret);
4332 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4334 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4336 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4338 if (!IS_ERR(inode) && root != sub_root) {
4339 down_read(&root->fs_info->cleanup_work_sem);
4340 if (!(inode->i_sb->s_flags & MS_RDONLY))
4341 ret = btrfs_orphan_cleanup(sub_root);
4342 up_read(&root->fs_info->cleanup_work_sem);
4344 inode = ERR_PTR(ret);
4350 static int btrfs_dentry_delete(const struct dentry *dentry)
4352 struct btrfs_root *root;
4353 struct inode *inode = dentry->d_inode;
4355 if (!inode && !IS_ROOT(dentry))
4356 inode = dentry->d_parent->d_inode;
4359 root = BTRFS_I(inode)->root;
4360 if (btrfs_root_refs(&root->root_item) == 0)
4363 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4369 static void btrfs_dentry_release(struct dentry *dentry)
4371 if (dentry->d_fsdata)
4372 kfree(dentry->d_fsdata);
4375 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4380 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4381 if (unlikely(d_need_lookup(dentry))) {
4382 spin_lock(&dentry->d_lock);
4383 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4384 spin_unlock(&dentry->d_lock);
4389 unsigned char btrfs_filetype_table[] = {
4390 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4393 static int btrfs_real_readdir(struct file *filp, void *dirent,
4396 struct inode *inode = filp->f_dentry->d_inode;
4397 struct btrfs_root *root = BTRFS_I(inode)->root;
4398 struct btrfs_item *item;
4399 struct btrfs_dir_item *di;
4400 struct btrfs_key key;
4401 struct btrfs_key found_key;
4402 struct btrfs_path *path;
4403 struct list_head ins_list;
4404 struct list_head del_list;
4406 struct extent_buffer *leaf;
4408 unsigned char d_type;
4413 int key_type = BTRFS_DIR_INDEX_KEY;
4417 int is_curr = 0; /* filp->f_pos points to the current index? */
4419 /* FIXME, use a real flag for deciding about the key type */
4420 if (root->fs_info->tree_root == root)
4421 key_type = BTRFS_DIR_ITEM_KEY;
4423 /* special case for "." */
4424 if (filp->f_pos == 0) {
4425 over = filldir(dirent, ".", 1,
4426 filp->f_pos, btrfs_ino(inode), DT_DIR);
4431 /* special case for .., just use the back ref */
4432 if (filp->f_pos == 1) {
4433 u64 pino = parent_ino(filp->f_path.dentry);
4434 over = filldir(dirent, "..", 2,
4435 filp->f_pos, pino, DT_DIR);
4440 path = btrfs_alloc_path();
4446 if (key_type == BTRFS_DIR_INDEX_KEY) {
4447 INIT_LIST_HEAD(&ins_list);
4448 INIT_LIST_HEAD(&del_list);
4449 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4452 btrfs_set_key_type(&key, key_type);
4453 key.offset = filp->f_pos;
4454 key.objectid = btrfs_ino(inode);
4456 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4461 leaf = path->nodes[0];
4462 slot = path->slots[0];
4463 if (slot >= btrfs_header_nritems(leaf)) {
4464 ret = btrfs_next_leaf(root, path);
4472 item = btrfs_item_nr(leaf, slot);
4473 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4475 if (found_key.objectid != key.objectid)
4477 if (btrfs_key_type(&found_key) != key_type)
4479 if (found_key.offset < filp->f_pos)
4481 if (key_type == BTRFS_DIR_INDEX_KEY &&
4482 btrfs_should_delete_dir_index(&del_list,
4486 filp->f_pos = found_key.offset;
4489 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4491 di_total = btrfs_item_size(leaf, item);
4493 while (di_cur < di_total) {
4494 struct btrfs_key location;
4496 if (verify_dir_item(root, leaf, di))
4499 name_len = btrfs_dir_name_len(leaf, di);
4500 if (name_len <= sizeof(tmp_name)) {
4501 name_ptr = tmp_name;
4503 name_ptr = kmalloc(name_len, GFP_NOFS);
4509 read_extent_buffer(leaf, name_ptr,
4510 (unsigned long)(di + 1), name_len);
4512 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4513 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4516 /* is this a reference to our own snapshot? If so
4519 * In contrast to old kernels, we insert the snapshot's
4520 * dir item and dir index after it has been created, so
4521 * we won't find a reference to our own snapshot. We
4522 * still keep the following code for backward
4525 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4526 location.objectid == root->root_key.objectid) {
4530 over = filldir(dirent, name_ptr, name_len,
4531 found_key.offset, location.objectid,
4535 if (name_ptr != tmp_name)
4540 di_len = btrfs_dir_name_len(leaf, di) +
4541 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4543 di = (struct btrfs_dir_item *)((char *)di + di_len);
4549 if (key_type == BTRFS_DIR_INDEX_KEY) {
4552 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4558 /* Reached end of directory/root. Bump pos past the last item. */
4559 if (key_type == BTRFS_DIR_INDEX_KEY)
4561 * 32-bit glibc will use getdents64, but then strtol -
4562 * so the last number we can serve is this.
4564 filp->f_pos = 0x7fffffff;
4570 if (key_type == BTRFS_DIR_INDEX_KEY)
4571 btrfs_put_delayed_items(&ins_list, &del_list);
4572 btrfs_free_path(path);
4576 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4578 struct btrfs_root *root = BTRFS_I(inode)->root;
4579 struct btrfs_trans_handle *trans;
4581 bool nolock = false;
4583 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4586 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4589 if (wbc->sync_mode == WB_SYNC_ALL) {
4591 trans = btrfs_join_transaction_nolock(root);
4593 trans = btrfs_join_transaction(root);
4595 return PTR_ERR(trans);
4596 ret = btrfs_commit_transaction(trans, root);
4602 * This is somewhat expensive, updating the tree every time the
4603 * inode changes. But, it is most likely to find the inode in cache.
4604 * FIXME, needs more benchmarking...there are no reasons other than performance
4605 * to keep or drop this code.
4607 int btrfs_dirty_inode(struct inode *inode)
4609 struct btrfs_root *root = BTRFS_I(inode)->root;
4610 struct btrfs_trans_handle *trans;
4613 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4616 trans = btrfs_join_transaction(root);
4618 return PTR_ERR(trans);
4620 ret = btrfs_update_inode(trans, root, inode);
4621 if (ret && ret == -ENOSPC) {
4622 /* whoops, lets try again with the full transaction */
4623 btrfs_end_transaction(trans, root);
4624 trans = btrfs_start_transaction(root, 1);
4626 return PTR_ERR(trans);
4628 ret = btrfs_update_inode(trans, root, inode);
4630 btrfs_end_transaction(trans, root);
4631 if (BTRFS_I(inode)->delayed_node)
4632 btrfs_balance_delayed_items(root);
4638 * This is a copy of file_update_time. We need this so we can return error on
4639 * ENOSPC for updating the inode in the case of file write and mmap writes.
4641 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4644 struct btrfs_root *root = BTRFS_I(inode)->root;
4646 if (btrfs_root_readonly(root))
4649 if (flags & S_VERSION)
4650 inode_inc_iversion(inode);
4651 if (flags & S_CTIME)
4652 inode->i_ctime = *now;
4653 if (flags & S_MTIME)
4654 inode->i_mtime = *now;
4655 if (flags & S_ATIME)
4656 inode->i_atime = *now;
4657 return btrfs_dirty_inode(inode);
4661 * find the highest existing sequence number in a directory
4662 * and then set the in-memory index_cnt variable to reflect
4663 * free sequence numbers
4665 static int btrfs_set_inode_index_count(struct inode *inode)
4667 struct btrfs_root *root = BTRFS_I(inode)->root;
4668 struct btrfs_key key, found_key;
4669 struct btrfs_path *path;
4670 struct extent_buffer *leaf;
4673 key.objectid = btrfs_ino(inode);
4674 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4675 key.offset = (u64)-1;
4677 path = btrfs_alloc_path();
4681 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4684 /* FIXME: we should be able to handle this */
4690 * MAGIC NUMBER EXPLANATION:
4691 * since we search a directory based on f_pos we have to start at 2
4692 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4693 * else has to start at 2
4695 if (path->slots[0] == 0) {
4696 BTRFS_I(inode)->index_cnt = 2;
4702 leaf = path->nodes[0];
4703 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4705 if (found_key.objectid != btrfs_ino(inode) ||
4706 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4707 BTRFS_I(inode)->index_cnt = 2;
4711 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4713 btrfs_free_path(path);
4718 * helper to find a free sequence number in a given directory. This current
4719 * code is very simple, later versions will do smarter things in the btree
4721 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4725 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4726 ret = btrfs_inode_delayed_dir_index_count(dir);
4728 ret = btrfs_set_inode_index_count(dir);
4734 *index = BTRFS_I(dir)->index_cnt;
4735 BTRFS_I(dir)->index_cnt++;
4740 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4741 struct btrfs_root *root,
4743 const char *name, int name_len,
4744 u64 ref_objectid, u64 objectid,
4745 umode_t mode, u64 *index)
4747 struct inode *inode;
4748 struct btrfs_inode_item *inode_item;
4749 struct btrfs_key *location;
4750 struct btrfs_path *path;
4751 struct btrfs_inode_ref *ref;
4752 struct btrfs_key key[2];
4758 path = btrfs_alloc_path();
4760 return ERR_PTR(-ENOMEM);
4762 inode = new_inode(root->fs_info->sb);
4764 btrfs_free_path(path);
4765 return ERR_PTR(-ENOMEM);
4769 * we have to initialize this early, so we can reclaim the inode
4770 * number if we fail afterwards in this function.
4772 inode->i_ino = objectid;
4775 trace_btrfs_inode_request(dir);
4777 ret = btrfs_set_inode_index(dir, index);
4779 btrfs_free_path(path);
4781 return ERR_PTR(ret);
4785 * index_cnt is ignored for everything but a dir,
4786 * btrfs_get_inode_index_count has an explanation for the magic
4789 BTRFS_I(inode)->index_cnt = 2;
4790 BTRFS_I(inode)->root = root;
4791 BTRFS_I(inode)->generation = trans->transid;
4792 inode->i_generation = BTRFS_I(inode)->generation;
4795 * We could have gotten an inode number from somebody who was fsynced
4796 * and then removed in this same transaction, so let's just set full
4797 * sync since it will be a full sync anyway and this will blow away the
4798 * old info in the log.
4800 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4807 key[0].objectid = objectid;
4808 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4812 * Start new inodes with an inode_ref. This is slightly more
4813 * efficient for small numbers of hard links since they will
4814 * be packed into one item. Extended refs will kick in if we
4815 * add more hard links than can fit in the ref item.
4817 key[1].objectid = objectid;
4818 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4819 key[1].offset = ref_objectid;
4821 sizes[0] = sizeof(struct btrfs_inode_item);
4822 sizes[1] = name_len + sizeof(*ref);
4824 path->leave_spinning = 1;
4825 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4829 inode_init_owner(inode, dir, mode);
4830 inode_set_bytes(inode, 0);
4831 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4832 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4833 struct btrfs_inode_item);
4834 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4835 sizeof(*inode_item));
4836 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4838 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4839 struct btrfs_inode_ref);
4840 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4841 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4842 ptr = (unsigned long)(ref + 1);
4843 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4845 btrfs_mark_buffer_dirty(path->nodes[0]);
4846 btrfs_free_path(path);
4848 location = &BTRFS_I(inode)->location;
4849 location->objectid = objectid;
4850 location->offset = 0;
4851 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4853 btrfs_inherit_iflags(inode, dir);
4855 if (S_ISREG(mode)) {
4856 if (btrfs_test_opt(root, NODATASUM))
4857 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4858 if (btrfs_test_opt(root, NODATACOW))
4859 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4862 insert_inode_hash(inode);
4863 inode_tree_add(inode);
4865 trace_btrfs_inode_new(inode);
4866 btrfs_set_inode_last_trans(trans, inode);
4868 btrfs_update_root_times(trans, root);
4873 BTRFS_I(dir)->index_cnt--;
4874 btrfs_free_path(path);
4876 return ERR_PTR(ret);
4879 static inline u8 btrfs_inode_type(struct inode *inode)
4881 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4885 * utility function to add 'inode' into 'parent_inode' with
4886 * a give name and a given sequence number.
4887 * if 'add_backref' is true, also insert a backref from the
4888 * inode to the parent directory.
4890 int btrfs_add_link(struct btrfs_trans_handle *trans,
4891 struct inode *parent_inode, struct inode *inode,
4892 const char *name, int name_len, int add_backref, u64 index)
4895 struct btrfs_key key;
4896 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4897 u64 ino = btrfs_ino(inode);
4898 u64 parent_ino = btrfs_ino(parent_inode);
4900 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4901 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4904 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4908 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4909 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4910 key.objectid, root->root_key.objectid,
4911 parent_ino, index, name, name_len);
4912 } else if (add_backref) {
4913 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4917 /* Nothing to clean up yet */
4921 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4923 btrfs_inode_type(inode), index);
4924 if (ret == -EEXIST || ret == -EOVERFLOW)
4927 btrfs_abort_transaction(trans, root, ret);
4931 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4933 inode_inc_iversion(parent_inode);
4934 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4935 ret = btrfs_update_inode(trans, root, parent_inode);
4937 btrfs_abort_transaction(trans, root, ret);
4941 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4944 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4945 key.objectid, root->root_key.objectid,
4946 parent_ino, &local_index, name, name_len);
4948 } else if (add_backref) {
4952 err = btrfs_del_inode_ref(trans, root, name, name_len,
4953 ino, parent_ino, &local_index);
4958 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4959 struct inode *dir, struct dentry *dentry,
4960 struct inode *inode, int backref, u64 index)
4962 int err = btrfs_add_link(trans, dir, inode,
4963 dentry->d_name.name, dentry->d_name.len,
4970 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4971 umode_t mode, dev_t rdev)
4973 struct btrfs_trans_handle *trans;
4974 struct btrfs_root *root = BTRFS_I(dir)->root;
4975 struct inode *inode = NULL;
4981 if (!new_valid_dev(rdev))
4985 * 2 for inode item and ref
4987 * 1 for xattr if selinux is on
4989 trans = btrfs_start_transaction(root, 5);
4991 return PTR_ERR(trans);
4993 err = btrfs_find_free_ino(root, &objectid);
4997 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4998 dentry->d_name.len, btrfs_ino(dir), objectid,
5000 if (IS_ERR(inode)) {
5001 err = PTR_ERR(inode);
5005 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5011 err = btrfs_update_inode(trans, root, inode);
5018 * If the active LSM wants to access the inode during
5019 * d_instantiate it needs these. Smack checks to see
5020 * if the filesystem supports xattrs by looking at the
5024 inode->i_op = &btrfs_special_inode_operations;
5025 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5029 init_special_inode(inode, inode->i_mode, rdev);
5030 btrfs_update_inode(trans, root, inode);
5031 d_instantiate(dentry, inode);
5034 btrfs_end_transaction(trans, root);
5035 btrfs_btree_balance_dirty(root);
5037 inode_dec_link_count(inode);
5043 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5044 umode_t mode, bool excl)
5046 struct btrfs_trans_handle *trans;
5047 struct btrfs_root *root = BTRFS_I(dir)->root;
5048 struct inode *inode = NULL;
5049 int drop_inode_on_err = 0;
5055 * 2 for inode item and ref
5057 * 1 for xattr if selinux is on
5059 trans = btrfs_start_transaction(root, 5);
5061 return PTR_ERR(trans);
5063 err = btrfs_find_free_ino(root, &objectid);
5067 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5068 dentry->d_name.len, btrfs_ino(dir), objectid,
5070 if (IS_ERR(inode)) {
5071 err = PTR_ERR(inode);
5074 drop_inode_on_err = 1;
5076 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5080 err = btrfs_update_inode(trans, root, inode);
5085 * If the active LSM wants to access the inode during
5086 * d_instantiate it needs these. Smack checks to see
5087 * if the filesystem supports xattrs by looking at the
5090 inode->i_fop = &btrfs_file_operations;
5091 inode->i_op = &btrfs_file_inode_operations;
5093 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5097 inode->i_mapping->a_ops = &btrfs_aops;
5098 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5099 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5100 d_instantiate(dentry, inode);
5103 btrfs_end_transaction(trans, root);
5104 if (err && drop_inode_on_err) {
5105 inode_dec_link_count(inode);
5108 btrfs_btree_balance_dirty(root);
5112 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5113 struct dentry *dentry)
5115 struct btrfs_trans_handle *trans;
5116 struct btrfs_root *root = BTRFS_I(dir)->root;
5117 struct inode *inode = old_dentry->d_inode;
5122 /* do not allow sys_link's with other subvols of the same device */
5123 if (root->objectid != BTRFS_I(inode)->root->objectid)
5126 if (inode->i_nlink >= BTRFS_LINK_MAX)
5129 err = btrfs_set_inode_index(dir, &index);
5134 * 2 items for inode and inode ref
5135 * 2 items for dir items
5136 * 1 item for parent inode
5138 trans = btrfs_start_transaction(root, 5);
5139 if (IS_ERR(trans)) {
5140 err = PTR_ERR(trans);
5144 btrfs_inc_nlink(inode);
5145 inode_inc_iversion(inode);
5146 inode->i_ctime = CURRENT_TIME;
5148 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5150 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5155 struct dentry *parent = dentry->d_parent;
5156 err = btrfs_update_inode(trans, root, inode);
5159 d_instantiate(dentry, inode);
5160 btrfs_log_new_name(trans, inode, NULL, parent);
5163 btrfs_end_transaction(trans, root);
5166 inode_dec_link_count(inode);
5169 btrfs_btree_balance_dirty(root);
5173 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5175 struct inode *inode = NULL;
5176 struct btrfs_trans_handle *trans;
5177 struct btrfs_root *root = BTRFS_I(dir)->root;
5179 int drop_on_err = 0;
5184 * 2 items for inode and ref
5185 * 2 items for dir items
5186 * 1 for xattr if selinux is on
5188 trans = btrfs_start_transaction(root, 5);
5190 return PTR_ERR(trans);
5192 err = btrfs_find_free_ino(root, &objectid);
5196 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5197 dentry->d_name.len, btrfs_ino(dir), objectid,
5198 S_IFDIR | mode, &index);
5199 if (IS_ERR(inode)) {
5200 err = PTR_ERR(inode);
5206 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5210 inode->i_op = &btrfs_dir_inode_operations;
5211 inode->i_fop = &btrfs_dir_file_operations;
5213 btrfs_i_size_write(inode, 0);
5214 err = btrfs_update_inode(trans, root, inode);
5218 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5219 dentry->d_name.len, 0, index);
5223 d_instantiate(dentry, inode);
5227 btrfs_end_transaction(trans, root);
5230 btrfs_btree_balance_dirty(root);
5234 /* helper for btfs_get_extent. Given an existing extent in the tree,
5235 * and an extent that you want to insert, deal with overlap and insert
5236 * the new extent into the tree.
5238 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5239 struct extent_map *existing,
5240 struct extent_map *em,
5241 u64 map_start, u64 map_len)
5245 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5246 start_diff = map_start - em->start;
5247 em->start = map_start;
5249 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5250 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5251 em->block_start += start_diff;
5252 em->block_len -= start_diff;
5254 return add_extent_mapping(em_tree, em);
5257 static noinline int uncompress_inline(struct btrfs_path *path,
5258 struct inode *inode, struct page *page,
5259 size_t pg_offset, u64 extent_offset,
5260 struct btrfs_file_extent_item *item)
5263 struct extent_buffer *leaf = path->nodes[0];
5266 unsigned long inline_size;
5270 WARN_ON(pg_offset != 0);
5271 compress_type = btrfs_file_extent_compression(leaf, item);
5272 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5273 inline_size = btrfs_file_extent_inline_item_len(leaf,
5274 btrfs_item_nr(leaf, path->slots[0]));
5275 tmp = kmalloc(inline_size, GFP_NOFS);
5278 ptr = btrfs_file_extent_inline_start(item);
5280 read_extent_buffer(leaf, tmp, ptr, inline_size);
5282 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5283 ret = btrfs_decompress(compress_type, tmp, page,
5284 extent_offset, inline_size, max_size);
5286 char *kaddr = kmap_atomic(page);
5287 unsigned long copy_size = min_t(u64,
5288 PAGE_CACHE_SIZE - pg_offset,
5289 max_size - extent_offset);
5290 memset(kaddr + pg_offset, 0, copy_size);
5291 kunmap_atomic(kaddr);
5298 * a bit scary, this does extent mapping from logical file offset to the disk.
5299 * the ugly parts come from merging extents from the disk with the in-ram
5300 * representation. This gets more complex because of the data=ordered code,
5301 * where the in-ram extents might be locked pending data=ordered completion.
5303 * This also copies inline extents directly into the page.
5306 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5307 size_t pg_offset, u64 start, u64 len,
5313 u64 extent_start = 0;
5315 u64 objectid = btrfs_ino(inode);
5317 struct btrfs_path *path = NULL;
5318 struct btrfs_root *root = BTRFS_I(inode)->root;
5319 struct btrfs_file_extent_item *item;
5320 struct extent_buffer *leaf;
5321 struct btrfs_key found_key;
5322 struct extent_map *em = NULL;
5323 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5324 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5325 struct btrfs_trans_handle *trans = NULL;
5329 read_lock(&em_tree->lock);
5330 em = lookup_extent_mapping(em_tree, start, len);
5332 em->bdev = root->fs_info->fs_devices->latest_bdev;
5333 read_unlock(&em_tree->lock);
5336 if (em->start > start || em->start + em->len <= start)
5337 free_extent_map(em);
5338 else if (em->block_start == EXTENT_MAP_INLINE && page)
5339 free_extent_map(em);
5343 em = alloc_extent_map();
5348 em->bdev = root->fs_info->fs_devices->latest_bdev;
5349 em->start = EXTENT_MAP_HOLE;
5350 em->orig_start = EXTENT_MAP_HOLE;
5352 em->block_len = (u64)-1;
5355 path = btrfs_alloc_path();
5361 * Chances are we'll be called again, so go ahead and do
5367 ret = btrfs_lookup_file_extent(trans, root, path,
5368 objectid, start, trans != NULL);
5375 if (path->slots[0] == 0)
5380 leaf = path->nodes[0];
5381 item = btrfs_item_ptr(leaf, path->slots[0],
5382 struct btrfs_file_extent_item);
5383 /* are we inside the extent that was found? */
5384 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5385 found_type = btrfs_key_type(&found_key);
5386 if (found_key.objectid != objectid ||
5387 found_type != BTRFS_EXTENT_DATA_KEY) {
5391 found_type = btrfs_file_extent_type(leaf, item);
5392 extent_start = found_key.offset;
5393 compress_type = btrfs_file_extent_compression(leaf, item);
5394 if (found_type == BTRFS_FILE_EXTENT_REG ||
5395 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5396 extent_end = extent_start +
5397 btrfs_file_extent_num_bytes(leaf, item);
5398 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5400 size = btrfs_file_extent_inline_len(leaf, item);
5401 extent_end = (extent_start + size + root->sectorsize - 1) &
5402 ~((u64)root->sectorsize - 1);
5405 if (start >= extent_end) {
5407 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5408 ret = btrfs_next_leaf(root, path);
5415 leaf = path->nodes[0];
5417 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5418 if (found_key.objectid != objectid ||
5419 found_key.type != BTRFS_EXTENT_DATA_KEY)
5421 if (start + len <= found_key.offset)
5424 em->orig_start = start;
5425 em->len = found_key.offset - start;
5429 if (found_type == BTRFS_FILE_EXTENT_REG ||
5430 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5431 em->start = extent_start;
5432 em->len = extent_end - extent_start;
5433 em->orig_start = extent_start -
5434 btrfs_file_extent_offset(leaf, item);
5435 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5437 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5439 em->block_start = EXTENT_MAP_HOLE;
5442 if (compress_type != BTRFS_COMPRESS_NONE) {
5443 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5444 em->compress_type = compress_type;
5445 em->block_start = bytenr;
5446 em->block_len = em->orig_block_len;
5448 bytenr += btrfs_file_extent_offset(leaf, item);
5449 em->block_start = bytenr;
5450 em->block_len = em->len;
5451 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5452 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5455 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5459 size_t extent_offset;
5462 em->block_start = EXTENT_MAP_INLINE;
5463 if (!page || create) {
5464 em->start = extent_start;
5465 em->len = extent_end - extent_start;
5469 size = btrfs_file_extent_inline_len(leaf, item);
5470 extent_offset = page_offset(page) + pg_offset - extent_start;
5471 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5472 size - extent_offset);
5473 em->start = extent_start + extent_offset;
5474 em->len = (copy_size + root->sectorsize - 1) &
5475 ~((u64)root->sectorsize - 1);
5476 em->orig_block_len = em->len;
5477 em->orig_start = em->start;
5478 if (compress_type) {
5479 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5480 em->compress_type = compress_type;
5482 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5483 if (create == 0 && !PageUptodate(page)) {
5484 if (btrfs_file_extent_compression(leaf, item) !=
5485 BTRFS_COMPRESS_NONE) {
5486 ret = uncompress_inline(path, inode, page,
5488 extent_offset, item);
5489 BUG_ON(ret); /* -ENOMEM */
5492 read_extent_buffer(leaf, map + pg_offset, ptr,
5494 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5495 memset(map + pg_offset + copy_size, 0,
5496 PAGE_CACHE_SIZE - pg_offset -
5501 flush_dcache_page(page);
5502 } else if (create && PageUptodate(page)) {
5506 free_extent_map(em);
5509 btrfs_release_path(path);
5510 trans = btrfs_join_transaction(root);
5513 return ERR_CAST(trans);
5517 write_extent_buffer(leaf, map + pg_offset, ptr,
5520 btrfs_mark_buffer_dirty(leaf);
5522 set_extent_uptodate(io_tree, em->start,
5523 extent_map_end(em) - 1, NULL, GFP_NOFS);
5526 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5530 em->orig_start = start;
5533 em->block_start = EXTENT_MAP_HOLE;
5534 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5536 btrfs_release_path(path);
5537 if (em->start > start || extent_map_end(em) <= start) {
5538 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5539 "[%llu %llu]\n", (unsigned long long)em->start,
5540 (unsigned long long)em->len,
5541 (unsigned long long)start,
5542 (unsigned long long)len);
5548 write_lock(&em_tree->lock);
5549 ret = add_extent_mapping(em_tree, em);
5550 /* it is possible that someone inserted the extent into the tree
5551 * while we had the lock dropped. It is also possible that
5552 * an overlapping map exists in the tree
5554 if (ret == -EEXIST) {
5555 struct extent_map *existing;
5559 existing = lookup_extent_mapping(em_tree, start, len);
5560 if (existing && (existing->start > start ||
5561 existing->start + existing->len <= start)) {
5562 free_extent_map(existing);
5566 existing = lookup_extent_mapping(em_tree, em->start,
5569 err = merge_extent_mapping(em_tree, existing,
5572 free_extent_map(existing);
5574 free_extent_map(em);
5579 free_extent_map(em);
5583 free_extent_map(em);
5588 write_unlock(&em_tree->lock);
5592 trace_btrfs_get_extent(root, em);
5595 btrfs_free_path(path);
5597 ret = btrfs_end_transaction(trans, root);
5602 free_extent_map(em);
5603 return ERR_PTR(err);
5605 BUG_ON(!em); /* Error is always set */
5609 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5610 size_t pg_offset, u64 start, u64 len,
5613 struct extent_map *em;
5614 struct extent_map *hole_em = NULL;
5615 u64 range_start = start;
5621 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5626 * if our em maps to a hole, there might
5627 * actually be delalloc bytes behind it
5629 if (em->block_start != EXTENT_MAP_HOLE)
5635 /* check to see if we've wrapped (len == -1 or similar) */
5644 /* ok, we didn't find anything, lets look for delalloc */
5645 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5646 end, len, EXTENT_DELALLOC, 1);
5647 found_end = range_start + found;
5648 if (found_end < range_start)
5649 found_end = (u64)-1;
5652 * we didn't find anything useful, return
5653 * the original results from get_extent()
5655 if (range_start > end || found_end <= start) {
5661 /* adjust the range_start to make sure it doesn't
5662 * go backwards from the start they passed in
5664 range_start = max(start,range_start);
5665 found = found_end - range_start;
5668 u64 hole_start = start;
5671 em = alloc_extent_map();
5677 * when btrfs_get_extent can't find anything it
5678 * returns one huge hole
5680 * make sure what it found really fits our range, and
5681 * adjust to make sure it is based on the start from
5685 u64 calc_end = extent_map_end(hole_em);
5687 if (calc_end <= start || (hole_em->start > end)) {
5688 free_extent_map(hole_em);
5691 hole_start = max(hole_em->start, start);
5692 hole_len = calc_end - hole_start;
5696 if (hole_em && range_start > hole_start) {
5697 /* our hole starts before our delalloc, so we
5698 * have to return just the parts of the hole
5699 * that go until the delalloc starts
5701 em->len = min(hole_len,
5702 range_start - hole_start);
5703 em->start = hole_start;
5704 em->orig_start = hole_start;
5706 * don't adjust block start at all,
5707 * it is fixed at EXTENT_MAP_HOLE
5709 em->block_start = hole_em->block_start;
5710 em->block_len = hole_len;
5712 em->start = range_start;
5714 em->orig_start = range_start;
5715 em->block_start = EXTENT_MAP_DELALLOC;
5716 em->block_len = found;
5718 } else if (hole_em) {
5723 free_extent_map(hole_em);
5725 free_extent_map(em);
5726 return ERR_PTR(err);
5731 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5734 struct btrfs_root *root = BTRFS_I(inode)->root;
5735 struct btrfs_trans_handle *trans;
5736 struct extent_map *em;
5737 struct btrfs_key ins;
5741 trans = btrfs_join_transaction(root);
5743 return ERR_CAST(trans);
5745 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5747 alloc_hint = get_extent_allocation_hint(inode, start, len);
5748 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5749 alloc_hint, &ins, 1);
5755 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5756 ins.offset, ins.offset, 0);
5760 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5761 ins.offset, ins.offset, 0);
5763 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5767 btrfs_end_transaction(trans, root);
5772 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5773 * block must be cow'd
5775 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5776 struct inode *inode, u64 offset, u64 len)
5778 struct btrfs_path *path;
5780 struct extent_buffer *leaf;
5781 struct btrfs_root *root = BTRFS_I(inode)->root;
5782 struct btrfs_file_extent_item *fi;
5783 struct btrfs_key key;
5791 path = btrfs_alloc_path();
5795 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5800 slot = path->slots[0];
5803 /* can't find the item, must cow */
5810 leaf = path->nodes[0];
5811 btrfs_item_key_to_cpu(leaf, &key, slot);
5812 if (key.objectid != btrfs_ino(inode) ||
5813 key.type != BTRFS_EXTENT_DATA_KEY) {
5814 /* not our file or wrong item type, must cow */
5818 if (key.offset > offset) {
5819 /* Wrong offset, must cow */
5823 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5824 found_type = btrfs_file_extent_type(leaf, fi);
5825 if (found_type != BTRFS_FILE_EXTENT_REG &&
5826 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5827 /* not a regular extent, must cow */
5830 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5831 backref_offset = btrfs_file_extent_offset(leaf, fi);
5833 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5834 if (extent_end < offset + len) {
5835 /* extent doesn't include our full range, must cow */
5839 if (btrfs_extent_readonly(root, disk_bytenr))
5843 * look for other files referencing this extent, if we
5844 * find any we must cow
5846 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5847 key.offset - backref_offset, disk_bytenr))
5851 * adjust disk_bytenr and num_bytes to cover just the bytes
5852 * in this extent we are about to write. If there
5853 * are any csums in that range we have to cow in order
5854 * to keep the csums correct
5856 disk_bytenr += backref_offset;
5857 disk_bytenr += offset - key.offset;
5858 num_bytes = min(offset + len, extent_end) - offset;
5859 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5862 * all of the above have passed, it is safe to overwrite this extent
5867 btrfs_free_path(path);
5871 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5872 struct extent_state **cached_state, int writing)
5874 struct btrfs_ordered_extent *ordered;
5878 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5881 * We're concerned with the entire range that we're going to be
5882 * doing DIO to, so we need to make sure theres no ordered
5883 * extents in this range.
5885 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5886 lockend - lockstart + 1);
5889 * We need to make sure there are no buffered pages in this
5890 * range either, we could have raced between the invalidate in
5891 * generic_file_direct_write and locking the extent. The
5892 * invalidate needs to happen so that reads after a write do not
5895 if (!ordered && (!writing ||
5896 !test_range_bit(&BTRFS_I(inode)->io_tree,
5897 lockstart, lockend, EXTENT_UPTODATE, 0,
5901 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5902 cached_state, GFP_NOFS);
5905 btrfs_start_ordered_extent(inode, ordered, 1);
5906 btrfs_put_ordered_extent(ordered);
5908 /* Screw you mmap */
5909 ret = filemap_write_and_wait_range(inode->i_mapping,
5916 * If we found a page that couldn't be invalidated just
5917 * fall back to buffered.
5919 ret = invalidate_inode_pages2_range(inode->i_mapping,
5920 lockstart >> PAGE_CACHE_SHIFT,
5921 lockend >> PAGE_CACHE_SHIFT);
5932 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5933 u64 len, u64 orig_start,
5934 u64 block_start, u64 block_len,
5935 u64 orig_block_len, int type)
5937 struct extent_map_tree *em_tree;
5938 struct extent_map *em;
5939 struct btrfs_root *root = BTRFS_I(inode)->root;
5942 em_tree = &BTRFS_I(inode)->extent_tree;
5943 em = alloc_extent_map();
5945 return ERR_PTR(-ENOMEM);
5948 em->orig_start = orig_start;
5950 em->block_len = block_len;
5951 em->block_start = block_start;
5952 em->bdev = root->fs_info->fs_devices->latest_bdev;
5953 em->orig_block_len = orig_block_len;
5954 em->generation = -1;
5955 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5956 if (type == BTRFS_ORDERED_PREALLOC)
5957 set_bit(EXTENT_FLAG_FILLING, &em->flags);
5960 btrfs_drop_extent_cache(inode, em->start,
5961 em->start + em->len - 1, 0);
5962 write_lock(&em_tree->lock);
5963 ret = add_extent_mapping(em_tree, em);
5965 list_move(&em->list,
5966 &em_tree->modified_extents);
5967 write_unlock(&em_tree->lock);
5968 } while (ret == -EEXIST);
5971 free_extent_map(em);
5972 return ERR_PTR(ret);
5979 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5980 struct buffer_head *bh_result, int create)
5982 struct extent_map *em;
5983 struct btrfs_root *root = BTRFS_I(inode)->root;
5984 struct extent_state *cached_state = NULL;
5985 u64 start = iblock << inode->i_blkbits;
5986 u64 lockstart, lockend;
5987 u64 len = bh_result->b_size;
5988 struct btrfs_trans_handle *trans;
5989 int unlock_bits = EXTENT_LOCKED;
5993 ret = btrfs_delalloc_reserve_space(inode, len);
5996 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5998 len = min_t(u64, len, root->sectorsize);
6002 lockend = start + len - 1;
6005 * If this errors out it's because we couldn't invalidate pagecache for
6006 * this range and we need to fallback to buffered.
6008 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6012 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6013 lockend, EXTENT_DELALLOC, NULL,
6014 &cached_state, GFP_NOFS);
6019 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6026 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6027 * io. INLINE is special, and we could probably kludge it in here, but
6028 * it's still buffered so for safety lets just fall back to the generic
6031 * For COMPRESSED we _have_ to read the entire extent in so we can
6032 * decompress it, so there will be buffering required no matter what we
6033 * do, so go ahead and fallback to buffered.
6035 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6036 * to buffered IO. Don't blame me, this is the price we pay for using
6039 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6040 em->block_start == EXTENT_MAP_INLINE) {
6041 free_extent_map(em);
6046 /* Just a good old fashioned hole, return */
6047 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6048 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6049 free_extent_map(em);
6055 * We don't allocate a new extent in the following cases
6057 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6059 * 2) The extent is marked as PREALLOC. We're good to go here and can
6060 * just use the extent.
6064 len = min(len, em->len - (start - em->start));
6065 lockstart = start + len;
6069 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6070 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6071 em->block_start != EXTENT_MAP_HOLE)) {
6076 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6077 type = BTRFS_ORDERED_PREALLOC;
6079 type = BTRFS_ORDERED_NOCOW;
6080 len = min(len, em->len - (start - em->start));
6081 block_start = em->block_start + (start - em->start);
6084 * we're not going to log anything, but we do need
6085 * to make sure the current transaction stays open
6086 * while we look for nocow cross refs
6088 trans = btrfs_join_transaction(root);
6092 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6093 u64 orig_start = em->orig_start;
6094 u64 orig_block_len = em->orig_block_len;
6096 if (type == BTRFS_ORDERED_PREALLOC) {
6097 free_extent_map(em);
6098 em = create_pinned_em(inode, start, len,
6101 orig_block_len, type);
6103 btrfs_end_transaction(trans, root);
6108 ret = btrfs_add_ordered_extent_dio(inode, start,
6109 block_start, len, len, type);
6110 btrfs_end_transaction(trans, root);
6112 free_extent_map(em);
6117 btrfs_end_transaction(trans, root);
6121 * this will cow the extent, reset the len in case we changed
6124 len = bh_result->b_size;
6125 free_extent_map(em);
6126 em = btrfs_new_extent_direct(inode, start, len);
6131 len = min(len, em->len - (start - em->start));
6133 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6135 bh_result->b_size = len;
6136 bh_result->b_bdev = em->bdev;
6137 set_buffer_mapped(bh_result);
6139 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6140 set_buffer_new(bh_result);
6143 * Need to update the i_size under the extent lock so buffered
6144 * readers will get the updated i_size when we unlock.
6146 if (start + len > i_size_read(inode))
6147 i_size_write(inode, start + len);
6151 * In the case of write we need to clear and unlock the entire range,
6152 * in the case of read we need to unlock only the end area that we
6153 * aren't using if there is any left over space.
6155 if (lockstart < lockend) {
6156 if (create && len < lockend - lockstart) {
6157 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6158 lockstart + len - 1,
6159 unlock_bits | EXTENT_DEFRAG, 1, 0,
6160 &cached_state, GFP_NOFS);
6162 * Beside unlock, we also need to cleanup reserved space
6163 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6165 clear_extent_bit(&BTRFS_I(inode)->io_tree,
6166 lockstart + len, lockend,
6167 unlock_bits | EXTENT_DO_ACCOUNTING |
6168 EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6170 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6171 lockend, unlock_bits, 1, 0,
6172 &cached_state, GFP_NOFS);
6175 free_extent_state(cached_state);
6178 free_extent_map(em);
6184 unlock_bits |= EXTENT_DO_ACCOUNTING;
6186 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6187 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6191 struct btrfs_dio_private {
6192 struct inode *inode;
6198 /* number of bios pending for this dio */
6199 atomic_t pending_bios;
6204 struct bio *orig_bio;
6207 static void btrfs_endio_direct_read(struct bio *bio, int err)
6209 struct btrfs_dio_private *dip = bio->bi_private;
6210 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6211 struct bio_vec *bvec = bio->bi_io_vec;
6212 struct inode *inode = dip->inode;
6213 struct btrfs_root *root = BTRFS_I(inode)->root;
6216 start = dip->logical_offset;
6218 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6219 struct page *page = bvec->bv_page;
6222 u64 private = ~(u32)0;
6223 unsigned long flags;
6225 if (get_state_private(&BTRFS_I(inode)->io_tree,
6228 local_irq_save(flags);
6229 kaddr = kmap_atomic(page);
6230 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6231 csum, bvec->bv_len);
6232 btrfs_csum_final(csum, (char *)&csum);
6233 kunmap_atomic(kaddr);
6234 local_irq_restore(flags);
6236 flush_dcache_page(bvec->bv_page);
6237 if (csum != private) {
6239 printk(KERN_ERR "btrfs csum failed ino %llu off"
6240 " %llu csum %u private %u\n",
6241 (unsigned long long)btrfs_ino(inode),
6242 (unsigned long long)start,
6243 csum, (unsigned)private);
6248 start += bvec->bv_len;
6250 } while (bvec <= bvec_end);
6252 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6253 dip->logical_offset + dip->bytes - 1);
6254 bio->bi_private = dip->private;
6258 /* If we had a csum failure make sure to clear the uptodate flag */
6260 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6261 dio_end_io(bio, err);
6264 static void btrfs_endio_direct_write(struct bio *bio, int err)
6266 struct btrfs_dio_private *dip = bio->bi_private;
6267 struct inode *inode = dip->inode;
6268 struct btrfs_root *root = BTRFS_I(inode)->root;
6269 struct btrfs_ordered_extent *ordered = NULL;
6270 u64 ordered_offset = dip->logical_offset;
6271 u64 ordered_bytes = dip->bytes;
6277 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6279 ordered_bytes, !err);
6283 ordered->work.func = finish_ordered_fn;
6284 ordered->work.flags = 0;
6285 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6289 * our bio might span multiple ordered extents. If we haven't
6290 * completed the accounting for the whole dio, go back and try again
6292 if (ordered_offset < dip->logical_offset + dip->bytes) {
6293 ordered_bytes = dip->logical_offset + dip->bytes -
6299 bio->bi_private = dip->private;
6303 /* If we had an error make sure to clear the uptodate flag */
6305 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6306 dio_end_io(bio, err);
6309 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6310 struct bio *bio, int mirror_num,
6311 unsigned long bio_flags, u64 offset)
6314 struct btrfs_root *root = BTRFS_I(inode)->root;
6315 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6316 BUG_ON(ret); /* -ENOMEM */
6320 static void btrfs_end_dio_bio(struct bio *bio, int err)
6322 struct btrfs_dio_private *dip = bio->bi_private;
6325 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6326 "sector %#Lx len %u err no %d\n",
6327 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6328 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6332 * before atomic variable goto zero, we must make sure
6333 * dip->errors is perceived to be set.
6335 smp_mb__before_atomic_dec();
6338 /* if there are more bios still pending for this dio, just exit */
6339 if (!atomic_dec_and_test(&dip->pending_bios))
6343 bio_io_error(dip->orig_bio);
6345 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6346 bio_endio(dip->orig_bio, 0);
6352 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6353 u64 first_sector, gfp_t gfp_flags)
6355 int nr_vecs = bio_get_nr_vecs(bdev);
6356 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6359 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6360 int rw, u64 file_offset, int skip_sum,
6363 int write = rw & REQ_WRITE;
6364 struct btrfs_root *root = BTRFS_I(inode)->root;
6368 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6373 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6381 if (write && async_submit) {
6382 ret = btrfs_wq_submit_bio(root->fs_info,
6383 inode, rw, bio, 0, 0,
6385 __btrfs_submit_bio_start_direct_io,
6386 __btrfs_submit_bio_done);
6390 * If we aren't doing async submit, calculate the csum of the
6393 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6396 } else if (!skip_sum) {
6397 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6403 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6409 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6412 struct inode *inode = dip->inode;
6413 struct btrfs_root *root = BTRFS_I(inode)->root;
6415 struct bio *orig_bio = dip->orig_bio;
6416 struct bio_vec *bvec = orig_bio->bi_io_vec;
6417 u64 start_sector = orig_bio->bi_sector;
6418 u64 file_offset = dip->logical_offset;
6423 int async_submit = 0;
6425 map_length = orig_bio->bi_size;
6426 ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6427 &map_length, NULL, 0);
6433 if (map_length >= orig_bio->bi_size) {
6439 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6442 bio->bi_private = dip;
6443 bio->bi_end_io = btrfs_end_dio_bio;
6444 atomic_inc(&dip->pending_bios);
6446 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6447 if (unlikely(map_length < submit_len + bvec->bv_len ||
6448 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6449 bvec->bv_offset) < bvec->bv_len)) {
6451 * inc the count before we submit the bio so
6452 * we know the end IO handler won't happen before
6453 * we inc the count. Otherwise, the dip might get freed
6454 * before we're done setting it up
6456 atomic_inc(&dip->pending_bios);
6457 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6458 file_offset, skip_sum,
6462 atomic_dec(&dip->pending_bios);
6466 start_sector += submit_len >> 9;
6467 file_offset += submit_len;
6472 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6473 start_sector, GFP_NOFS);
6476 bio->bi_private = dip;
6477 bio->bi_end_io = btrfs_end_dio_bio;
6479 map_length = orig_bio->bi_size;
6480 ret = btrfs_map_block(root->fs_info, READ,
6482 &map_length, NULL, 0);
6488 submit_len += bvec->bv_len;
6495 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6504 * before atomic variable goto zero, we must
6505 * make sure dip->errors is perceived to be set.
6507 smp_mb__before_atomic_dec();
6508 if (atomic_dec_and_test(&dip->pending_bios))
6509 bio_io_error(dip->orig_bio);
6511 /* bio_end_io() will handle error, so we needn't return it */
6515 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6518 struct btrfs_root *root = BTRFS_I(inode)->root;
6519 struct btrfs_dio_private *dip;
6520 struct bio_vec *bvec = bio->bi_io_vec;
6522 int write = rw & REQ_WRITE;
6525 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6527 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6533 dip->private = bio->bi_private;
6535 dip->logical_offset = file_offset;
6539 dip->bytes += bvec->bv_len;
6541 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6543 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6544 bio->bi_private = dip;
6546 dip->orig_bio = bio;
6547 atomic_set(&dip->pending_bios, 0);
6550 bio->bi_end_io = btrfs_endio_direct_write;
6552 bio->bi_end_io = btrfs_endio_direct_read;
6554 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6559 * If this is a write, we need to clean up the reserved space and kill
6560 * the ordered extent.
6563 struct btrfs_ordered_extent *ordered;
6564 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6565 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6566 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6567 btrfs_free_reserved_extent(root, ordered->start,
6569 btrfs_put_ordered_extent(ordered);
6570 btrfs_put_ordered_extent(ordered);
6572 bio_endio(bio, ret);
6575 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6576 const struct iovec *iov, loff_t offset,
6577 unsigned long nr_segs)
6583 unsigned blocksize_mask = root->sectorsize - 1;
6584 ssize_t retval = -EINVAL;
6585 loff_t end = offset;
6587 if (offset & blocksize_mask)
6590 /* Check the memory alignment. Blocks cannot straddle pages */
6591 for (seg = 0; seg < nr_segs; seg++) {
6592 addr = (unsigned long)iov[seg].iov_base;
6593 size = iov[seg].iov_len;
6595 if ((addr & blocksize_mask) || (size & blocksize_mask))
6598 /* If this is a write we don't need to check anymore */
6603 * Check to make sure we don't have duplicate iov_base's in this
6604 * iovec, if so return EINVAL, otherwise we'll get csum errors
6605 * when reading back.
6607 for (i = seg + 1; i < nr_segs; i++) {
6608 if (iov[seg].iov_base == iov[i].iov_base)
6617 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6618 const struct iovec *iov, loff_t offset,
6619 unsigned long nr_segs)
6621 struct file *file = iocb->ki_filp;
6622 struct inode *inode = file->f_mapping->host;
6624 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6628 return __blockdev_direct_IO(rw, iocb, inode,
6629 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6630 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6631 btrfs_submit_direct, 0);
6634 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
6636 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6637 __u64 start, __u64 len)
6641 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6645 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6648 int btrfs_readpage(struct file *file, struct page *page)
6650 struct extent_io_tree *tree;
6651 tree = &BTRFS_I(page->mapping->host)->io_tree;
6652 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6655 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6657 struct extent_io_tree *tree;
6660 if (current->flags & PF_MEMALLOC) {
6661 redirty_page_for_writepage(wbc, page);
6665 tree = &BTRFS_I(page->mapping->host)->io_tree;
6666 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6669 int btrfs_writepages(struct address_space *mapping,
6670 struct writeback_control *wbc)
6672 struct extent_io_tree *tree;
6674 tree = &BTRFS_I(mapping->host)->io_tree;
6675 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6679 btrfs_readpages(struct file *file, struct address_space *mapping,
6680 struct list_head *pages, unsigned nr_pages)
6682 struct extent_io_tree *tree;
6683 tree = &BTRFS_I(mapping->host)->io_tree;
6684 return extent_readpages(tree, mapping, pages, nr_pages,
6687 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6689 struct extent_io_tree *tree;
6690 struct extent_map_tree *map;
6693 tree = &BTRFS_I(page->mapping->host)->io_tree;
6694 map = &BTRFS_I(page->mapping->host)->extent_tree;
6695 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6697 ClearPagePrivate(page);
6698 set_page_private(page, 0);
6699 page_cache_release(page);
6704 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6706 if (PageWriteback(page) || PageDirty(page))
6708 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6711 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6713 struct inode *inode = page->mapping->host;
6714 struct extent_io_tree *tree;
6715 struct btrfs_ordered_extent *ordered;
6716 struct extent_state *cached_state = NULL;
6717 u64 page_start = page_offset(page);
6718 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6721 * we have the page locked, so new writeback can't start,
6722 * and the dirty bit won't be cleared while we are here.
6724 * Wait for IO on this page so that we can safely clear
6725 * the PagePrivate2 bit and do ordered accounting
6727 wait_on_page_writeback(page);
6729 tree = &BTRFS_I(inode)->io_tree;
6731 btrfs_releasepage(page, GFP_NOFS);
6734 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6735 ordered = btrfs_lookup_ordered_extent(inode,
6739 * IO on this page will never be started, so we need
6740 * to account for any ordered extents now
6742 clear_extent_bit(tree, page_start, page_end,
6743 EXTENT_DIRTY | EXTENT_DELALLOC |
6744 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6745 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6747 * whoever cleared the private bit is responsible
6748 * for the finish_ordered_io
6750 if (TestClearPagePrivate2(page) &&
6751 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6752 PAGE_CACHE_SIZE, 1)) {
6753 btrfs_finish_ordered_io(ordered);
6755 btrfs_put_ordered_extent(ordered);
6756 cached_state = NULL;
6757 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6759 clear_extent_bit(tree, page_start, page_end,
6760 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6761 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6762 &cached_state, GFP_NOFS);
6763 __btrfs_releasepage(page, GFP_NOFS);
6765 ClearPageChecked(page);
6766 if (PagePrivate(page)) {
6767 ClearPagePrivate(page);
6768 set_page_private(page, 0);
6769 page_cache_release(page);
6774 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6775 * called from a page fault handler when a page is first dirtied. Hence we must
6776 * be careful to check for EOF conditions here. We set the page up correctly
6777 * for a written page which means we get ENOSPC checking when writing into
6778 * holes and correct delalloc and unwritten extent mapping on filesystems that
6779 * support these features.
6781 * We are not allowed to take the i_mutex here so we have to play games to
6782 * protect against truncate races as the page could now be beyond EOF. Because
6783 * vmtruncate() writes the inode size before removing pages, once we have the
6784 * page lock we can determine safely if the page is beyond EOF. If it is not
6785 * beyond EOF, then the page is guaranteed safe against truncation until we
6788 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6790 struct page *page = vmf->page;
6791 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6792 struct btrfs_root *root = BTRFS_I(inode)->root;
6793 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6794 struct btrfs_ordered_extent *ordered;
6795 struct extent_state *cached_state = NULL;
6797 unsigned long zero_start;
6804 sb_start_pagefault(inode->i_sb);
6805 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6807 ret = file_update_time(vma->vm_file);
6813 else /* -ENOSPC, -EIO, etc */
6814 ret = VM_FAULT_SIGBUS;
6820 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6823 size = i_size_read(inode);
6824 page_start = page_offset(page);
6825 page_end = page_start + PAGE_CACHE_SIZE - 1;
6827 if ((page->mapping != inode->i_mapping) ||
6828 (page_start >= size)) {
6829 /* page got truncated out from underneath us */
6832 wait_on_page_writeback(page);
6834 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6835 set_page_extent_mapped(page);
6838 * we can't set the delalloc bits if there are pending ordered
6839 * extents. Drop our locks and wait for them to finish
6841 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6843 unlock_extent_cached(io_tree, page_start, page_end,
6844 &cached_state, GFP_NOFS);
6846 btrfs_start_ordered_extent(inode, ordered, 1);
6847 btrfs_put_ordered_extent(ordered);
6852 * XXX - page_mkwrite gets called every time the page is dirtied, even
6853 * if it was already dirty, so for space accounting reasons we need to
6854 * clear any delalloc bits for the range we are fixing to save. There
6855 * is probably a better way to do this, but for now keep consistent with
6856 * prepare_pages in the normal write path.
6858 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6859 EXTENT_DIRTY | EXTENT_DELALLOC |
6860 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6861 0, 0, &cached_state, GFP_NOFS);
6863 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6866 unlock_extent_cached(io_tree, page_start, page_end,
6867 &cached_state, GFP_NOFS);
6868 ret = VM_FAULT_SIGBUS;
6873 /* page is wholly or partially inside EOF */
6874 if (page_start + PAGE_CACHE_SIZE > size)
6875 zero_start = size & ~PAGE_CACHE_MASK;
6877 zero_start = PAGE_CACHE_SIZE;
6879 if (zero_start != PAGE_CACHE_SIZE) {
6881 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6882 flush_dcache_page(page);
6885 ClearPageChecked(page);
6886 set_page_dirty(page);
6887 SetPageUptodate(page);
6889 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6890 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6891 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6893 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6897 sb_end_pagefault(inode->i_sb);
6898 return VM_FAULT_LOCKED;
6902 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6904 sb_end_pagefault(inode->i_sb);
6908 static int btrfs_truncate(struct inode *inode)
6910 struct btrfs_root *root = BTRFS_I(inode)->root;
6911 struct btrfs_block_rsv *rsv;
6914 struct btrfs_trans_handle *trans;
6915 u64 mask = root->sectorsize - 1;
6916 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6918 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6922 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6923 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6926 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6927 * 3 things going on here
6929 * 1) We need to reserve space for our orphan item and the space to
6930 * delete our orphan item. Lord knows we don't want to have a dangling
6931 * orphan item because we didn't reserve space to remove it.
6933 * 2) We need to reserve space to update our inode.
6935 * 3) We need to have something to cache all the space that is going to
6936 * be free'd up by the truncate operation, but also have some slack
6937 * space reserved in case it uses space during the truncate (thank you
6938 * very much snapshotting).
6940 * And we need these to all be seperate. The fact is we can use alot of
6941 * space doing the truncate, and we have no earthly idea how much space
6942 * we will use, so we need the truncate reservation to be seperate so it
6943 * doesn't end up using space reserved for updating the inode or
6944 * removing the orphan item. We also need to be able to stop the
6945 * transaction and start a new one, which means we need to be able to
6946 * update the inode several times, and we have no idea of knowing how
6947 * many times that will be, so we can't just reserve 1 item for the
6948 * entirety of the opration, so that has to be done seperately as well.
6949 * Then there is the orphan item, which does indeed need to be held on
6950 * to for the whole operation, and we need nobody to touch this reserved
6951 * space except the orphan code.
6953 * So that leaves us with
6955 * 1) root->orphan_block_rsv - for the orphan deletion.
6956 * 2) rsv - for the truncate reservation, which we will steal from the
6957 * transaction reservation.
6958 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6959 * updating the inode.
6961 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6964 rsv->size = min_size;
6968 * 1 for the truncate slack space
6969 * 1 for updating the inode.
6971 trans = btrfs_start_transaction(root, 2);
6972 if (IS_ERR(trans)) {
6973 err = PTR_ERR(trans);
6977 /* Migrate the slack space for the truncate to our reserve */
6978 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6983 * setattr is responsible for setting the ordered_data_close flag,
6984 * but that is only tested during the last file release. That
6985 * could happen well after the next commit, leaving a great big
6986 * window where new writes may get lost if someone chooses to write
6987 * to this file after truncating to zero
6989 * The inode doesn't have any dirty data here, and so if we commit
6990 * this is a noop. If someone immediately starts writing to the inode
6991 * it is very likely we'll catch some of their writes in this
6992 * transaction, and the commit will find this file on the ordered
6993 * data list with good things to send down.
6995 * This is a best effort solution, there is still a window where
6996 * using truncate to replace the contents of the file will
6997 * end up with a zero length file after a crash.
6999 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7000 &BTRFS_I(inode)->runtime_flags))
7001 btrfs_add_ordered_operation(trans, root, inode);
7004 * So if we truncate and then write and fsync we normally would just
7005 * write the extents that changed, which is a problem if we need to
7006 * first truncate that entire inode. So set this flag so we write out
7007 * all of the extents in the inode to the sync log so we're completely
7010 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7011 trans->block_rsv = rsv;
7014 ret = btrfs_truncate_inode_items(trans, root, inode,
7016 BTRFS_EXTENT_DATA_KEY);
7017 if (ret != -ENOSPC) {
7022 trans->block_rsv = &root->fs_info->trans_block_rsv;
7023 ret = btrfs_update_inode(trans, root, inode);
7029 btrfs_end_transaction(trans, root);
7030 btrfs_btree_balance_dirty(root);
7032 trans = btrfs_start_transaction(root, 2);
7033 if (IS_ERR(trans)) {
7034 ret = err = PTR_ERR(trans);
7039 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7041 BUG_ON(ret); /* shouldn't happen */
7042 trans->block_rsv = rsv;
7045 if (ret == 0 && inode->i_nlink > 0) {
7046 trans->block_rsv = root->orphan_block_rsv;
7047 ret = btrfs_orphan_del(trans, inode);
7053 trans->block_rsv = &root->fs_info->trans_block_rsv;
7054 ret = btrfs_update_inode(trans, root, inode);
7058 ret = btrfs_end_transaction(trans, root);
7059 btrfs_btree_balance_dirty(root);
7063 btrfs_free_block_rsv(root, rsv);
7072 * create a new subvolume directory/inode (helper for the ioctl).
7074 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7075 struct btrfs_root *new_root, u64 new_dirid)
7077 struct inode *inode;
7081 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7082 new_dirid, new_dirid,
7083 S_IFDIR | (~current_umask() & S_IRWXUGO),
7086 return PTR_ERR(inode);
7087 inode->i_op = &btrfs_dir_inode_operations;
7088 inode->i_fop = &btrfs_dir_file_operations;
7090 set_nlink(inode, 1);
7091 btrfs_i_size_write(inode, 0);
7093 err = btrfs_update_inode(trans, new_root, inode);
7099 struct inode *btrfs_alloc_inode(struct super_block *sb)
7101 struct btrfs_inode *ei;
7102 struct inode *inode;
7104 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7111 ei->last_sub_trans = 0;
7112 ei->logged_trans = 0;
7113 ei->delalloc_bytes = 0;
7114 ei->disk_i_size = 0;
7117 ei->index_cnt = (u64)-1;
7118 ei->last_unlink_trans = 0;
7119 ei->last_log_commit = 0;
7121 spin_lock_init(&ei->lock);
7122 ei->outstanding_extents = 0;
7123 ei->reserved_extents = 0;
7125 ei->runtime_flags = 0;
7126 ei->force_compress = BTRFS_COMPRESS_NONE;
7128 ei->delayed_node = NULL;
7130 inode = &ei->vfs_inode;
7131 extent_map_tree_init(&ei->extent_tree);
7132 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7133 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7134 ei->io_tree.track_uptodate = 1;
7135 ei->io_failure_tree.track_uptodate = 1;
7136 atomic_set(&ei->sync_writers, 0);
7137 mutex_init(&ei->log_mutex);
7138 mutex_init(&ei->delalloc_mutex);
7139 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7140 INIT_LIST_HEAD(&ei->delalloc_inodes);
7141 INIT_LIST_HEAD(&ei->ordered_operations);
7142 RB_CLEAR_NODE(&ei->rb_node);
7147 static void btrfs_i_callback(struct rcu_head *head)
7149 struct inode *inode = container_of(head, struct inode, i_rcu);
7150 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7153 void btrfs_destroy_inode(struct inode *inode)
7155 struct btrfs_ordered_extent *ordered;
7156 struct btrfs_root *root = BTRFS_I(inode)->root;
7158 WARN_ON(!hlist_empty(&inode->i_dentry));
7159 WARN_ON(inode->i_data.nrpages);
7160 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7161 WARN_ON(BTRFS_I(inode)->reserved_extents);
7162 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7163 WARN_ON(BTRFS_I(inode)->csum_bytes);
7166 * This can happen where we create an inode, but somebody else also
7167 * created the same inode and we need to destroy the one we already
7174 * Make sure we're properly removed from the ordered operation
7178 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7179 spin_lock(&root->fs_info->ordered_extent_lock);
7180 list_del_init(&BTRFS_I(inode)->ordered_operations);
7181 spin_unlock(&root->fs_info->ordered_extent_lock);
7184 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7185 &BTRFS_I(inode)->runtime_flags)) {
7186 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7187 (unsigned long long)btrfs_ino(inode));
7188 atomic_dec(&root->orphan_inodes);
7192 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7196 printk(KERN_ERR "btrfs found ordered "
7197 "extent %llu %llu on inode cleanup\n",
7198 (unsigned long long)ordered->file_offset,
7199 (unsigned long long)ordered->len);
7200 btrfs_remove_ordered_extent(inode, ordered);
7201 btrfs_put_ordered_extent(ordered);
7202 btrfs_put_ordered_extent(ordered);
7205 inode_tree_del(inode);
7206 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7208 btrfs_remove_delayed_node(inode);
7209 call_rcu(&inode->i_rcu, btrfs_i_callback);
7212 int btrfs_drop_inode(struct inode *inode)
7214 struct btrfs_root *root = BTRFS_I(inode)->root;
7216 if (btrfs_root_refs(&root->root_item) == 0 &&
7217 !btrfs_is_free_space_inode(inode))
7220 return generic_drop_inode(inode);
7223 static void init_once(void *foo)
7225 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7227 inode_init_once(&ei->vfs_inode);
7230 void btrfs_destroy_cachep(void)
7233 * Make sure all delayed rcu free inodes are flushed before we
7237 if (btrfs_inode_cachep)
7238 kmem_cache_destroy(btrfs_inode_cachep);
7239 if (btrfs_trans_handle_cachep)
7240 kmem_cache_destroy(btrfs_trans_handle_cachep);
7241 if (btrfs_transaction_cachep)
7242 kmem_cache_destroy(btrfs_transaction_cachep);
7243 if (btrfs_path_cachep)
7244 kmem_cache_destroy(btrfs_path_cachep);
7245 if (btrfs_free_space_cachep)
7246 kmem_cache_destroy(btrfs_free_space_cachep);
7247 if (btrfs_delalloc_work_cachep)
7248 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7251 int btrfs_init_cachep(void)
7253 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7254 sizeof(struct btrfs_inode), 0,
7255 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7256 if (!btrfs_inode_cachep)
7259 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7260 sizeof(struct btrfs_trans_handle), 0,
7261 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7262 if (!btrfs_trans_handle_cachep)
7265 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7266 sizeof(struct btrfs_transaction), 0,
7267 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7268 if (!btrfs_transaction_cachep)
7271 btrfs_path_cachep = kmem_cache_create("btrfs_path",
7272 sizeof(struct btrfs_path), 0,
7273 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7274 if (!btrfs_path_cachep)
7277 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7278 sizeof(struct btrfs_free_space), 0,
7279 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7280 if (!btrfs_free_space_cachep)
7283 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7284 sizeof(struct btrfs_delalloc_work), 0,
7285 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7287 if (!btrfs_delalloc_work_cachep)
7292 btrfs_destroy_cachep();
7296 static int btrfs_getattr(struct vfsmount *mnt,
7297 struct dentry *dentry, struct kstat *stat)
7299 struct inode *inode = dentry->d_inode;
7300 u32 blocksize = inode->i_sb->s_blocksize;
7302 generic_fillattr(inode, stat);
7303 stat->dev = BTRFS_I(inode)->root->anon_dev;
7304 stat->blksize = PAGE_CACHE_SIZE;
7305 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7306 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7311 * If a file is moved, it will inherit the cow and compression flags of the new
7314 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7316 struct btrfs_inode *b_dir = BTRFS_I(dir);
7317 struct btrfs_inode *b_inode = BTRFS_I(inode);
7319 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7320 b_inode->flags |= BTRFS_INODE_NODATACOW;
7322 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7324 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7325 b_inode->flags |= BTRFS_INODE_COMPRESS;
7326 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7328 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7329 BTRFS_INODE_NOCOMPRESS);
7333 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7334 struct inode *new_dir, struct dentry *new_dentry)
7336 struct btrfs_trans_handle *trans;
7337 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7338 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7339 struct inode *new_inode = new_dentry->d_inode;
7340 struct inode *old_inode = old_dentry->d_inode;
7341 struct timespec ctime = CURRENT_TIME;
7345 u64 old_ino = btrfs_ino(old_inode);
7347 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7350 /* we only allow rename subvolume link between subvolumes */
7351 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7354 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7355 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7358 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7359 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7363 /* check for collisions, even if the name isn't there */
7364 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7365 new_dentry->d_name.name,
7366 new_dentry->d_name.len);
7369 if (ret == -EEXIST) {
7371 * eexist without a new_inode */
7377 /* maybe -EOVERFLOW */
7384 * we're using rename to replace one file with another.
7385 * and the replacement file is large. Start IO on it now so
7386 * we don't add too much work to the end of the transaction
7388 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7389 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7390 filemap_flush(old_inode->i_mapping);
7392 /* close the racy window with snapshot create/destroy ioctl */
7393 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7394 down_read(&root->fs_info->subvol_sem);
7396 * We want to reserve the absolute worst case amount of items. So if
7397 * both inodes are subvols and we need to unlink them then that would
7398 * require 4 item modifications, but if they are both normal inodes it
7399 * would require 5 item modifications, so we'll assume their normal
7400 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7401 * should cover the worst case number of items we'll modify.
7403 trans = btrfs_start_transaction(root, 20);
7404 if (IS_ERR(trans)) {
7405 ret = PTR_ERR(trans);
7410 btrfs_record_root_in_trans(trans, dest);
7412 ret = btrfs_set_inode_index(new_dir, &index);
7416 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7417 /* force full log commit if subvolume involved. */
7418 root->fs_info->last_trans_log_full_commit = trans->transid;
7420 ret = btrfs_insert_inode_ref(trans, dest,
7421 new_dentry->d_name.name,
7422 new_dentry->d_name.len,
7424 btrfs_ino(new_dir), index);
7428 * this is an ugly little race, but the rename is required
7429 * to make sure that if we crash, the inode is either at the
7430 * old name or the new one. pinning the log transaction lets
7431 * us make sure we don't allow a log commit to come in after
7432 * we unlink the name but before we add the new name back in.
7434 btrfs_pin_log_trans(root);
7437 * make sure the inode gets flushed if it is replacing
7440 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7441 btrfs_add_ordered_operation(trans, root, old_inode);
7443 inode_inc_iversion(old_dir);
7444 inode_inc_iversion(new_dir);
7445 inode_inc_iversion(old_inode);
7446 old_dir->i_ctime = old_dir->i_mtime = ctime;
7447 new_dir->i_ctime = new_dir->i_mtime = ctime;
7448 old_inode->i_ctime = ctime;
7450 if (old_dentry->d_parent != new_dentry->d_parent)
7451 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7453 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7454 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7455 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7456 old_dentry->d_name.name,
7457 old_dentry->d_name.len);
7459 ret = __btrfs_unlink_inode(trans, root, old_dir,
7460 old_dentry->d_inode,
7461 old_dentry->d_name.name,
7462 old_dentry->d_name.len);
7464 ret = btrfs_update_inode(trans, root, old_inode);
7467 btrfs_abort_transaction(trans, root, ret);
7472 inode_inc_iversion(new_inode);
7473 new_inode->i_ctime = CURRENT_TIME;
7474 if (unlikely(btrfs_ino(new_inode) ==
7475 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7476 root_objectid = BTRFS_I(new_inode)->location.objectid;
7477 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7479 new_dentry->d_name.name,
7480 new_dentry->d_name.len);
7481 BUG_ON(new_inode->i_nlink == 0);
7483 ret = btrfs_unlink_inode(trans, dest, new_dir,
7484 new_dentry->d_inode,
7485 new_dentry->d_name.name,
7486 new_dentry->d_name.len);
7488 if (!ret && new_inode->i_nlink == 0) {
7489 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7493 btrfs_abort_transaction(trans, root, ret);
7498 fixup_inode_flags(new_dir, old_inode);
7500 ret = btrfs_add_link(trans, new_dir, old_inode,
7501 new_dentry->d_name.name,
7502 new_dentry->d_name.len, 0, index);
7504 btrfs_abort_transaction(trans, root, ret);
7508 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7509 struct dentry *parent = new_dentry->d_parent;
7510 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7511 btrfs_end_log_trans(root);
7514 btrfs_end_transaction(trans, root);
7516 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7517 up_read(&root->fs_info->subvol_sem);
7522 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7524 struct btrfs_delalloc_work *delalloc_work;
7526 delalloc_work = container_of(work, struct btrfs_delalloc_work,
7528 if (delalloc_work->wait)
7529 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7531 filemap_flush(delalloc_work->inode->i_mapping);
7533 if (delalloc_work->delay_iput)
7534 btrfs_add_delayed_iput(delalloc_work->inode);
7536 iput(delalloc_work->inode);
7537 complete(&delalloc_work->completion);
7540 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7541 int wait, int delay_iput)
7543 struct btrfs_delalloc_work *work;
7545 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7549 init_completion(&work->completion);
7550 INIT_LIST_HEAD(&work->list);
7551 work->inode = inode;
7553 work->delay_iput = delay_iput;
7554 work->work.func = btrfs_run_delalloc_work;
7559 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7561 wait_for_completion(&work->completion);
7562 kmem_cache_free(btrfs_delalloc_work_cachep, work);
7566 * some fairly slow code that needs optimization. This walks the list
7567 * of all the inodes with pending delalloc and forces them to disk.
7569 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7571 struct list_head *head = &root->fs_info->delalloc_inodes;
7572 struct btrfs_inode *binode;
7573 struct inode *inode;
7574 struct btrfs_delalloc_work *work, *next;
7575 struct list_head works;
7578 if (root->fs_info->sb->s_flags & MS_RDONLY)
7581 INIT_LIST_HEAD(&works);
7583 spin_lock(&root->fs_info->delalloc_lock);
7584 while (!list_empty(head)) {
7585 binode = list_entry(head->next, struct btrfs_inode,
7587 inode = igrab(&binode->vfs_inode);
7589 list_del_init(&binode->delalloc_inodes);
7590 spin_unlock(&root->fs_info->delalloc_lock);
7592 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7597 list_add_tail(&work->list, &works);
7598 btrfs_queue_worker(&root->fs_info->flush_workers,
7602 spin_lock(&root->fs_info->delalloc_lock);
7604 spin_unlock(&root->fs_info->delalloc_lock);
7606 /* the filemap_flush will queue IO into the worker threads, but
7607 * we have to make sure the IO is actually started and that
7608 * ordered extents get created before we return
7610 atomic_inc(&root->fs_info->async_submit_draining);
7611 while (atomic_read(&root->fs_info->nr_async_submits) ||
7612 atomic_read(&root->fs_info->async_delalloc_pages)) {
7613 wait_event(root->fs_info->async_submit_wait,
7614 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7615 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7617 atomic_dec(&root->fs_info->async_submit_draining);
7619 list_for_each_entry_safe(work, next, &works, list) {
7620 list_del_init(&work->list);
7621 btrfs_wait_and_free_delalloc_work(work);
7626 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7627 const char *symname)
7629 struct btrfs_trans_handle *trans;
7630 struct btrfs_root *root = BTRFS_I(dir)->root;
7631 struct btrfs_path *path;
7632 struct btrfs_key key;
7633 struct inode *inode = NULL;
7641 struct btrfs_file_extent_item *ei;
7642 struct extent_buffer *leaf;
7644 name_len = strlen(symname) + 1;
7645 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7646 return -ENAMETOOLONG;
7649 * 2 items for inode item and ref
7650 * 2 items for dir items
7651 * 1 item for xattr if selinux is on
7653 trans = btrfs_start_transaction(root, 5);
7655 return PTR_ERR(trans);
7657 err = btrfs_find_free_ino(root, &objectid);
7661 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7662 dentry->d_name.len, btrfs_ino(dir), objectid,
7663 S_IFLNK|S_IRWXUGO, &index);
7664 if (IS_ERR(inode)) {
7665 err = PTR_ERR(inode);
7669 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7676 * If the active LSM wants to access the inode during
7677 * d_instantiate it needs these. Smack checks to see
7678 * if the filesystem supports xattrs by looking at the
7681 inode->i_fop = &btrfs_file_operations;
7682 inode->i_op = &btrfs_file_inode_operations;
7684 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7688 inode->i_mapping->a_ops = &btrfs_aops;
7689 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7690 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7695 path = btrfs_alloc_path();
7701 key.objectid = btrfs_ino(inode);
7703 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7704 datasize = btrfs_file_extent_calc_inline_size(name_len);
7705 err = btrfs_insert_empty_item(trans, root, path, &key,
7709 btrfs_free_path(path);
7712 leaf = path->nodes[0];
7713 ei = btrfs_item_ptr(leaf, path->slots[0],
7714 struct btrfs_file_extent_item);
7715 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7716 btrfs_set_file_extent_type(leaf, ei,
7717 BTRFS_FILE_EXTENT_INLINE);
7718 btrfs_set_file_extent_encryption(leaf, ei, 0);
7719 btrfs_set_file_extent_compression(leaf, ei, 0);
7720 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7721 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7723 ptr = btrfs_file_extent_inline_start(ei);
7724 write_extent_buffer(leaf, symname, ptr, name_len);
7725 btrfs_mark_buffer_dirty(leaf);
7726 btrfs_free_path(path);
7728 inode->i_op = &btrfs_symlink_inode_operations;
7729 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7730 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7731 inode_set_bytes(inode, name_len);
7732 btrfs_i_size_write(inode, name_len - 1);
7733 err = btrfs_update_inode(trans, root, inode);
7739 d_instantiate(dentry, inode);
7740 btrfs_end_transaction(trans, root);
7742 inode_dec_link_count(inode);
7745 btrfs_btree_balance_dirty(root);
7749 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7750 u64 start, u64 num_bytes, u64 min_size,
7751 loff_t actual_len, u64 *alloc_hint,
7752 struct btrfs_trans_handle *trans)
7754 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7755 struct extent_map *em;
7756 struct btrfs_root *root = BTRFS_I(inode)->root;
7757 struct btrfs_key ins;
7758 u64 cur_offset = start;
7761 bool own_trans = true;
7765 while (num_bytes > 0) {
7767 trans = btrfs_start_transaction(root, 3);
7768 if (IS_ERR(trans)) {
7769 ret = PTR_ERR(trans);
7774 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7775 0, *alloc_hint, &ins, 1);
7778 btrfs_end_transaction(trans, root);
7782 ret = insert_reserved_file_extent(trans, inode,
7783 cur_offset, ins.objectid,
7784 ins.offset, ins.offset,
7785 ins.offset, 0, 0, 0,
7786 BTRFS_FILE_EXTENT_PREALLOC);
7788 btrfs_abort_transaction(trans, root, ret);
7790 btrfs_end_transaction(trans, root);
7793 btrfs_drop_extent_cache(inode, cur_offset,
7794 cur_offset + ins.offset -1, 0);
7796 em = alloc_extent_map();
7798 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7799 &BTRFS_I(inode)->runtime_flags);
7803 em->start = cur_offset;
7804 em->orig_start = cur_offset;
7805 em->len = ins.offset;
7806 em->block_start = ins.objectid;
7807 em->block_len = ins.offset;
7808 em->orig_block_len = ins.offset;
7809 em->bdev = root->fs_info->fs_devices->latest_bdev;
7810 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7811 em->generation = trans->transid;
7814 write_lock(&em_tree->lock);
7815 ret = add_extent_mapping(em_tree, em);
7817 list_move(&em->list,
7818 &em_tree->modified_extents);
7819 write_unlock(&em_tree->lock);
7822 btrfs_drop_extent_cache(inode, cur_offset,
7823 cur_offset + ins.offset - 1,
7826 free_extent_map(em);
7828 num_bytes -= ins.offset;
7829 cur_offset += ins.offset;
7830 *alloc_hint = ins.objectid + ins.offset;
7832 inode_inc_iversion(inode);
7833 inode->i_ctime = CURRENT_TIME;
7834 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7835 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7836 (actual_len > inode->i_size) &&
7837 (cur_offset > inode->i_size)) {
7838 if (cur_offset > actual_len)
7839 i_size = actual_len;
7841 i_size = cur_offset;
7842 i_size_write(inode, i_size);
7843 btrfs_ordered_update_i_size(inode, i_size, NULL);
7846 ret = btrfs_update_inode(trans, root, inode);
7849 btrfs_abort_transaction(trans, root, ret);
7851 btrfs_end_transaction(trans, root);
7856 btrfs_end_transaction(trans, root);
7861 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7862 u64 start, u64 num_bytes, u64 min_size,
7863 loff_t actual_len, u64 *alloc_hint)
7865 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7866 min_size, actual_len, alloc_hint,
7870 int btrfs_prealloc_file_range_trans(struct inode *inode,
7871 struct btrfs_trans_handle *trans, int mode,
7872 u64 start, u64 num_bytes, u64 min_size,
7873 loff_t actual_len, u64 *alloc_hint)
7875 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7876 min_size, actual_len, alloc_hint, trans);
7879 static int btrfs_set_page_dirty(struct page *page)
7881 return __set_page_dirty_nobuffers(page);
7884 static int btrfs_permission(struct inode *inode, int mask)
7886 struct btrfs_root *root = BTRFS_I(inode)->root;
7887 umode_t mode = inode->i_mode;
7889 if (mask & MAY_WRITE &&
7890 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7891 if (btrfs_root_readonly(root))
7893 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7896 return generic_permission(inode, mask);
7899 static const struct inode_operations btrfs_dir_inode_operations = {
7900 .getattr = btrfs_getattr,
7901 .lookup = btrfs_lookup,
7902 .create = btrfs_create,
7903 .unlink = btrfs_unlink,
7905 .mkdir = btrfs_mkdir,
7906 .rmdir = btrfs_rmdir,
7907 .rename = btrfs_rename,
7908 .symlink = btrfs_symlink,
7909 .setattr = btrfs_setattr,
7910 .mknod = btrfs_mknod,
7911 .setxattr = btrfs_setxattr,
7912 .getxattr = btrfs_getxattr,
7913 .listxattr = btrfs_listxattr,
7914 .removexattr = btrfs_removexattr,
7915 .permission = btrfs_permission,
7916 .get_acl = btrfs_get_acl,
7918 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7919 .lookup = btrfs_lookup,
7920 .permission = btrfs_permission,
7921 .get_acl = btrfs_get_acl,
7924 static const struct file_operations btrfs_dir_file_operations = {
7925 .llseek = generic_file_llseek,
7926 .read = generic_read_dir,
7927 .readdir = btrfs_real_readdir,
7928 .unlocked_ioctl = btrfs_ioctl,
7929 #ifdef CONFIG_COMPAT
7930 .compat_ioctl = btrfs_ioctl,
7932 .release = btrfs_release_file,
7933 .fsync = btrfs_sync_file,
7936 static struct extent_io_ops btrfs_extent_io_ops = {
7937 .fill_delalloc = run_delalloc_range,
7938 .submit_bio_hook = btrfs_submit_bio_hook,
7939 .merge_bio_hook = btrfs_merge_bio_hook,
7940 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7941 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7942 .writepage_start_hook = btrfs_writepage_start_hook,
7943 .set_bit_hook = btrfs_set_bit_hook,
7944 .clear_bit_hook = btrfs_clear_bit_hook,
7945 .merge_extent_hook = btrfs_merge_extent_hook,
7946 .split_extent_hook = btrfs_split_extent_hook,
7950 * btrfs doesn't support the bmap operation because swapfiles
7951 * use bmap to make a mapping of extents in the file. They assume
7952 * these extents won't change over the life of the file and they
7953 * use the bmap result to do IO directly to the drive.
7955 * the btrfs bmap call would return logical addresses that aren't
7956 * suitable for IO and they also will change frequently as COW
7957 * operations happen. So, swapfile + btrfs == corruption.
7959 * For now we're avoiding this by dropping bmap.
7961 static const struct address_space_operations btrfs_aops = {
7962 .readpage = btrfs_readpage,
7963 .writepage = btrfs_writepage,
7964 .writepages = btrfs_writepages,
7965 .readpages = btrfs_readpages,
7966 .direct_IO = btrfs_direct_IO,
7967 .invalidatepage = btrfs_invalidatepage,
7968 .releasepage = btrfs_releasepage,
7969 .set_page_dirty = btrfs_set_page_dirty,
7970 .error_remove_page = generic_error_remove_page,
7973 static const struct address_space_operations btrfs_symlink_aops = {
7974 .readpage = btrfs_readpage,
7975 .writepage = btrfs_writepage,
7976 .invalidatepage = btrfs_invalidatepage,
7977 .releasepage = btrfs_releasepage,
7980 static const struct inode_operations btrfs_file_inode_operations = {
7981 .getattr = btrfs_getattr,
7982 .setattr = btrfs_setattr,
7983 .setxattr = btrfs_setxattr,
7984 .getxattr = btrfs_getxattr,
7985 .listxattr = btrfs_listxattr,
7986 .removexattr = btrfs_removexattr,
7987 .permission = btrfs_permission,
7988 .fiemap = btrfs_fiemap,
7989 .get_acl = btrfs_get_acl,
7990 .update_time = btrfs_update_time,
7992 static const struct inode_operations btrfs_special_inode_operations = {
7993 .getattr = btrfs_getattr,
7994 .setattr = btrfs_setattr,
7995 .permission = btrfs_permission,
7996 .setxattr = btrfs_setxattr,
7997 .getxattr = btrfs_getxattr,
7998 .listxattr = btrfs_listxattr,
7999 .removexattr = btrfs_removexattr,
8000 .get_acl = btrfs_get_acl,
8001 .update_time = btrfs_update_time,
8003 static const struct inode_operations btrfs_symlink_inode_operations = {
8004 .readlink = generic_readlink,
8005 .follow_link = page_follow_link_light,
8006 .put_link = page_put_link,
8007 .getattr = btrfs_getattr,
8008 .setattr = btrfs_setattr,
8009 .permission = btrfs_permission,
8010 .setxattr = btrfs_setxattr,
8011 .getxattr = btrfs_getxattr,
8012 .listxattr = btrfs_listxattr,
8013 .removexattr = btrfs_removexattr,
8014 .get_acl = btrfs_get_acl,
8015 .update_time = btrfs_update_time,
8018 const struct dentry_operations btrfs_dentry_operations = {
8019 .d_delete = btrfs_dentry_delete,
8020 .d_release = btrfs_dentry_release,